US20210374109A1 - Apparatus, systems, and methods for batch and realtime data processing - Google Patents

Abstract

A traditional data processing system is configured to process input data either in batch or in real-time. On one hand, a batch data processing system is limiting because the batch data processing often cannot take into account any data received during the batch data processing. On the other hand, a real-time data processing system is limiting because the real-time system often cannot scale. The real-time data processing system is often limited to dealing with primitive data types and/or a small amount of data. Therefore, it is desirable to address the limitations of the batch data processing system and the real-time data processing system by combining the benefits of the batch data processing system and the real-time data processing system into a single data processing system.

Description

CROSS REFERENCE TO RELATED APPLICATIONS
• This application claims benefit of the earlier filing date, under 35 U.S.C. § 119(e), of
• • U.S. Provisional Application No. 61/799,986, filed on Mar. 15, 2013, entitled “SYSTEM FOR ANALYZING AND USING LOCATION BASED BEHAVIOR”;
  • U.S. Provisional Application No. 61/800,036, filed on Mar. 15, 2013, entitled “GEOGRAPHIC LOCATION DESCRIPTOR AND LINKER”;
  • U.S. Provisional Application No. 61/799,131, filed on Mar. 15, 2013, entitled “SYSTEM AND METHOD FOR CROWD SOURCING DOMAIN SPECIFIC INTELLIGENCE”;
  • U.S. Provisional Application No. 61/799,846, filed Mar. 15, 2013, entitled “SYSTEM WITH BATCH AND REAL TIME DATA PROCESSING”; and
  • U.S. Provisional Application No. 61/799,817, filed on Mar. 15, 2013, entitled “SYSTEM FOR ASSIGNING SCORES TO LOCATION ENTITIES”.
• This application is also related to:
• • U.S. patent application Ser. No. ______, entitled “APPARATUS, SYSTEMS, AND METHODS FOR ANALYZING MOVEMENTS OF TARGET ENTITIES,” identified by the Attorney Docket Number 2203957-00123US2, filed on the even-date herewith;
  • U.S. patent application Ser. No. ______, entitled “APPARATUS, SYSTEMS, AND METHODS FOR PROVIDING LOCATION INFORMATION,” identified by the Attorney Docket Number 2203957-00124US2, filed on the even-date herewith;
  • U.S. patent application Ser. No. ______, entitled “APPARATUS, SYSTEMS, AND METHODS FOR CROWDSOURCING DOMAIN SPECIFIC INTELLIGENCE,” identified by the Attorney Docket Number 2203957-00125US2, filed on the even-date herewith;
  • U.S. patent application Ser. No. ______, entitled “APPARATUS, SYSTEMS, AND METHODS FOR ANALYZING CHARACTERISTICS OF ENTITIES OF INTEREST,” identified by the Attorney Docket Number 2203957-00127US2, filed on the even-date herewith; and
  • U.S. patent application Ser. No. ______, entitled “APPARATUS, SYSTEMS, AND METHODS FOR GROUPING DATA RECORDS,” identified by the Attorney Docket Number 2203957-00129US1, filed on the even-date herewith.
• The entire content of each of the above-referenced applications (including both the provisional applications and the non-provisional applications) is herein incorporated by reference.
FIELD OF THE INVENTION
• The present disclosure generally relates to data processing systems, and specifically, to data processing systems that can process data using batch processing and real-time processing.
BACKGROUND
• The system disclosed herein relates to receiving, processing and storing data from many sources, representing the most “correct” summary of facts and opinions from the data, including being able to re-compute this in real-time, and then using the results to respond to queries. As an example, when a user inputs a query to a web-based system, mobile phone, or vehicle navigation system searching for a “child friendly Chinese restaurant in Greenwich Village that has valet parking”, the system can very quickly respond with a list of restaurants matching, for example, the attributes: {“kid_friendly”:true,“category”:“Restaurant>Chinese”, “valet_parking”:true, “neighborhood”:“Greenwich Village”}. A mobile phone may then provide a button to call each restaurant. The information describing each restaurant may be spread across many websites, sourced from many data stores, and provided directly by users of the system.
• A problem in the art is that all web pages, references, and data about all known businesses in the United States stored in any data store can be so large as to not be understandable and query-able in real-time. Updating and maintaining such a large amount of information can be difficult. For example, information describing businesses in the United States has more than billions of rows of input data, tens of billions of facts, and tens of terabytes of web content.
• At the same time, new information is continuously becoming available and it is desirable to include such information in the production of query results. As an example, the system may learn that a restaurant no longer offers valet parking, that the restaurant disallows children, or that the restaurant's phone number has been disconnected.
• Accordingly, it is desirable to be able to update a system that produces search results both on an ongoing basis (e.g., to account for newly written reviews) as well as on a whole-sale basis (e.g., to reevaluate the entire data and use information contained within that may have been previously unusable).
SUMMARY OF THE DISCLOSED SUBJECT MATTER
• The disclosed system is configured to receive new data (e.g., real-time updates) and old data (e.g., data that has been gathered over a long period of time) and is configured to periodically update the system based on a reevaluation of the new data and the old data.
• Unlike systems that can only search web pages and return links to matching web pages, the disclosed system can maintain attribute information about each entity and return the information directly. In conventional systems, a web search for “Chinese restaurant with valet parking”, for example, can return links to web pages with the words “Chinese”, “restaurant”, “valet”, and “parking”. This will generally include pages with statements like “there was no valet parking to be found” because the words “valet” and “parking” appeared in the text and were therefore indexed as keywords for the web page. By contrast, the disclosed system has attributes such as the category of the restaurant, and a value indicating whether the restaurant offers valet parking, which advantageously allow the system to respond with more meaningful results. Additionally, in the disclosed system, a user can operate as a contributor to correct the data. Also, the disclosed system can interpret facts across many web pages and arrive at a consensus answer, which is then query-able to further improve results.
• In embodiments of the disclosed system, a user can operate as a contributor that contributes data to the system. For example, a user may provide direct feedback to the system to correct a fact. Multiple such submissions can be considered together by the disclosed system along with information on websites such as blogs about child friendly restaurants and summarized into a rapidly evolving data store that can quickly respond to queries. Therefore, users of the disclosed system can access the newly corrected phone number and a more accurate assessment of its child-friendliness.
• In some embodiments, the disclosed system can improve or expand the analytic methods for understanding information on web pages or feedback. The analytic methods can be improved or expanded on an ongoing basis. For example, today's methods may be capable of extracting more information from a data source compared to the last month's methods. If one web page includes facts as simple text while another has opinions in complex prose, the disclosed system, using the last month's method, may have been able to process simple text data such as “Valet Parking: yes” but have been unable to process prose such as “There was no place to park, not even valet.” However, the disclosed system, using the today's method, may have expanded capability and be able to process the more nuanced prose data.
• In general, in an aspect, embodiments of the disclosed subject matter can include a computing system for generating a summary data of a set of data. The computing system can include one or more processors configured to run one or more modules stored in non-tangible computer readable medium. The one or more modules are operable to receive a first set of data and a second set of data, wherein the first set of data comprises a larger number of data items compared to the second set of data, process the first set of data to format the first set of data into a first structured set of data, generate a first summary data using the first structured set of data by operating rules for summarizing the first structured set of data, and store the first summary data in a data store, process the second set of data to format the second set of data into a second structured set of data, generate a second summary data based on the first structured set of data and the second structured set of data by operating rules for summarizing the first structured set of data and the second structured set of data, determine a difference between the first summary data and the second summary data, and update the data store based on the difference between the first summary data and the second summary data.
• In general, in an aspect, embodiments of the disclosed subject matter can include a method for generating a summary data of a set of data. The method can include receiving, at an input module operating on a processor of a computing system, a first set of data and a second set of data, wherein the first set of data comprises a larger number of data items compared to the second set of data, processing, at a first input processing module of the computing system, the first set of data to format the first set of data into a first structured set of data, generating, at a first summary generation module of the computing system, a first summary data using the first structured set of data by operating rules for summarizing the first structured set of data, maintaining the first summary data in a data store in the computing system, processing, at a second input processing module of the computing system, the second set of data to format the second set of data into a second structured set of data, generating, at a second summary generation module of the computing system, a second summary data using the first structured set of data and the second structured set of data by operating rules for summarizing the first structured set of data and the second structured set of data, determining, at a difference generation module of the computing system, a difference between the first summary data and the second summary data, and updating, by the computing system, the data store based on the difference between the first summary data and the second summary data.
• In general, in an aspect, embodiments of the disclosed subject matter can include a computer program product, tangibly embodied in a non-transitory computer-readable storage medium. The computer program product includes instructions operable to cause a data processing system to receive a first set of data and a second set of data, wherein the first set of data comprises a larger number of data items compared to the second set of data, process the first set of data to format the first set of data into a first structured set of data, generate a first summary data using the first structured set of data by operating rules for summarizing the first structured set of data, and store the first summary data in a data store, process the second set of data to format the second set of data into a second structured set of data, generate a second summary data using the first structured set of data and the second structured set of data by operating rules for summarizing the first structured set of data and the second structured set of data, determine a difference between the first summary data and the second summary data, and update the data store based on the difference between the first summary data and the second summary data.
• In any one of the embodiments disclosed herein, the second set of data comprises real-time data submissions, and the one or more modules are operable to process the second set of data to format the second set of data into the second structured set of data in response to receiving the second set of data.
• In any one of the embodiments disclosed herein, the computing system, the method, or the computer program product can include modules, steps, or executable instructions for processing the first set of data to format the first set of data into the first structured set of data at a first time interval, which is substantially longer than a second time interval at which the second set of data is formatted into the second structured set of data.
• In any one of the embodiments disclosed herein, each of the first summary data and the second summary data comprises an entity identifier and a value associated with the entity identifier, and wherein the computing system, the method, or the computer program product can further include modules, steps, or executable instructions for determining the difference between the first summary data and the second summary data by determining that the first summary data and the second summary data include an identical entity identifier, and comparing values associated with the identical entity identifiers in the first summary data and the second summary data.
• In any one of the embodiments disclosed herein, the computing system, the method, or the computer program product can include modules, steps, or executable instructions for providing the difference between the first summary data and the second summary data to other authorized computing systems.
• In any one of the embodiments disclosed herein, the computing system, the method, or the computer program product can include modules, steps, or executable instructions for providing the difference to other authorized computing systems via an application programming interface.
• In any one of the embodiments disclosed herein, the computing system, the method, or the computer program product can include modules, steps, or executable instructions for providing the difference to other authorized computing systems as a file.
• In any one of the embodiments disclosed herein, the computing system, the method, or the computer program product can include modules, steps, or executable instructions for combining at least the first set of data and the second set of data to generate a third set of data, processing the third set of data to format the third set of data into a third structured set of data based on new rules for formatting a set of data, and generating a third summary data using the third structured set of data.
• In any one of the embodiments disclosed herein, the first set of data and the third set of data each includes a first data element, and wherein the first data element is associated with a first entity in the first summary data identified by the first entity identifier, wherein the first data element is associated with a second entity in the third summary data, and wherein the computing system, the method, or the computer program product can further include modules, steps, or executable instructions for associating the first entity identifier to the second entity in the third summary data so that the first data element maintains its association with the first entity identifier in the third summary data.
• In any one of the embodiments disclosed herein, the first structured set of data comprises a grouping of data items based on an entity identifier associated with the data items.
• In any one of the embodiments disclosed herein, the computing system comprises at least one server in a data center.
• In any one of the embodiments disclosed herein, the data store comprises a plurality of data store systems, each of which is associated with a view, and wherein the one or more modules are operable to select one of the plurality of data store systems in response to a query based on the view associated with the query.
• In any one of the embodiments disclosed herein, the computing system, the method, or the computer program product can include modules, steps, or executable instructions for identifying a third set of data received after the generation of the second summary data, generating a third summary data based on the third set of data, the first structured set of data, and the second structured set of data by operating rules for summarizing the first structured set of data, the second structured set of data, and the third summary data, determining a difference between the second summary data and the third summary data, and updating the data store based on the difference between the second summary data and the third summary data.
DESCRIPTION OF THE FIGURES
• Various objects, features, and advantages of the present disclosure can be more fully appreciated with reference to the following detailed description when considered in connection with the following drawings, in which like reference numerals identify like elements. The following drawings are for the purpose of illustration only and are not intended to be limiting of the disclosed subject matter, the scope of which is set forth in the claims that follow.
• FIG. 1 illustrates the common processing framework of the disclosed system in accordance with some embodiments.
• FIGS. 2A-2C illustrate enlarged views of portions ofFIG. 1.
• FIG. 3 illustrates the Catchup process in accordance with some embodiments.
DETAILED DESCRIPTIONS
• A traditional data processing system is configured to process input data either in batch or in real-time. On one hand, a batch data processing system is limiting because the batch data processing cannot take into account any additional data received during the batch data processing. On the other hand, a real-time data processing system is limiting because the real-time system cannot scale. The real-time data processing system is often limited to dealing with primitive data types and/or a small amount of data. Therefore, it is desirable to address the limitations of the batch data processing system and the real-time data processing system by combining the benefits of the batch data processing system and the real-time data processing system into a single system.
• It is hard for a system to accommodate both a real-time processing and a batch processing because the data and/or processes for a real-time processing and a batch processing are quite different. For example, in a batch processing system, a program cannot access the data processing result until the entire data process is complete, whereas in a real-time processing system, a program can access the processing result during the data processing.
• The disclosed data processing apparatus, systems, and methods can address the challenges in integrating the batch data processing system and the real-time processing system.
• Some embodiments of the disclosed system can be configured to process unstructured data and convert the unstructured data into a summary data. The summary data can be stored in one or more data stores, including, for example, one or more data storages and/or one or more databases, or one or more search servers and can be formatted and optionally indexed to be query-able using one or more data stores or one or more search servers or using an application programming interface (API) by a third party user.
• The summary data can include one or more unique entities and at least one attribute about those entities. One of the attributes about an entity can be an entity identifier that is unique amongst the entities. Additional attributes describe some properties of the entity such as a Boolean value (e.g. whether restaurant A has valet parking is “True” or “False”), an integer, a string, a set of characters, binary data (e.g. bytes representing an image), or arrays or sets of these types, or any other combinations thereof.
• Some embodiments of the disclosed system can be configured to generate the summary data based on two types of data inputs: a bulk data input and an intermittent data input. The bulk data input can refer to a large amount of data that has been gathered over time. In some cases, the bulk data input can refer to all data that the disclosed system has received over a predetermined period of time, which can be long. For example, the bulk data input can include raw information received from multiple contributors or from a web-crawler over a long period of time. In some embodiments, the bulk data can be maintained in the disclosed system itself; in other embodiments, the bulk data can be received from another storage center viva communication interface. The intermittent data input can include a small amount of data that is provided to the disclosed system. The intermittent data input can include, for example, real-time data submissions from contributors.
• Some embodiments of the disclosed system can be configured to process both types of data inputs using a common processing framework. The common processing framework can include a real-time system that can respond to the intermittent data input (e.g., small incremental contributions from contributors) and reflect changes based on those contributions in consideration along with data from the batch system in the summary data in substantially real-time. The common professing framework can also include a batch system that can process the bulk data input. The batch system can be configured to format the bulk data to be amenable for further processing, and use the formatted bulk data to generate summary data.
• In some embodiments, the batch processing system is configured to generate summary data by formatting the unstructured data in the bulk data inputs into a structured data. Then the batch system is configured to group the elements in the structured data and generate a representative identifier for each group, also referred to as an entity. The batch system can then generate an identifier for each entity and calculate attribute values describing each entity.
• For example, when the large bulk data input includes 5 data elements associated with an existence of valet parking at a restaurant A, then the batch system can determine that those 5 data elements belong to the same entity (e.g., restaurant A), and consolidate information associated with the 5 data elements. For instance, if 3 data elements indicate that the restaurant A has valet parking and 2 data elements indicate that the restaurant A does not have valet parking, then the batch system can consolidate the 5 elements and indicate that the attribute “valet parking” for entity, restaurant A is “True.” This consolidation process is, in some ways, similar to the process disclosed in U.S. Patent Application Publication No. 2011/0066605, entitled “PROCESSES AND SYSTEMS FOR COLLABORATIVE MANIPULATION OF DATA,” filed on Sep. 15, 2009, which is herein incorporated by reference in its entirety.
• In some embodiments, an entity is a distinct object that the system elects to track. For example, the system can consider each physical restaurant (e.g. a chain with multiple locations would have an entity for each location) as a separate entity (i.e. summary record) when the system receives reviews about each physical restaurant as data input. Similarly, when toothpaste from a particular brand comes in 3 sizes and 4 flavors for each size, the system can maintain 12 distinct entities for the toothpaste.
• In some embodiments, the real-time system can be configured to update the summary data generated by the batch system as the real-time system receives intermittent data inputs from contributors. For example, if the real-time system receives two additional data inputs from the contributors, both indicating that the restaurant A does not have valet parking, then the real-time system can update the summary data to indicate that the attribute “valet parking” for entity “restaurant A” is “False.”
• In some embodiments, the real-time system can be configured to leverage the structured data generated by the batch system. For example, when the real-time system receives intermittent data input from a contributor, the real-time system can consolidate the intermittent data input with the structured bulk data generated by the batch system. This way, the amount of computation required by the real-time system can be reduced.
• In some embodiments, the batch system can be configured to run periodically, with a predetermined period. The batch system can be operated less frequently compared to the real-time system since the amount of computation needed by the batch system is considerably larger compared to the amount of computation needed by the real-time system. For example, the batch system can be operated so as to update the system on schedules like once an hour, once a week, or once a month. The real-time system can be operated more frequently than the batch system. For example, the real-time system can be configured to operate whenever the real-time system receives an intermittent data input, or on inputs buffered over a short time frame such as 5 seconds or 5 minutes. The batch system can be updated with new intelligence and rules over time and can process new data provided at a scale that is beyond the capacity of the real-time system.
• FIG. 1 illustrates the common processing framework of the disclosed system in accordance with some embodiments.FIGS. 2A-2C illustrate enlarged views of portions ofFIG. 1. The top portion ofFIG. 1 illustrates the processing performed by the real-time data system, whereas the bottom portion ofFIG. 1 illustrates the processing performed by the batch system.
Data Lifecycle
• One aspect of the disclosed data processing system is the data lifecycle. A data can be categorized as one of following types as the data progresses through the disclosed data processing system: Raw/Unprocessed Data100 (seeFIG. 2A), Unprocessed (Raw) Inputs350 (seeFIG. 2B), QuickProcessed Inputs150 (seeFIG. 2B), QuickProcessed Summaries190 (seeFIG. 2C), FullProcessed Inputs360 (seeFIG. 2B), and FullProcessed Summaries760 (seeFIG. 2C). These data can be stored in a non-transitory computer readable medium. The non-transitory computer readable medium can include one or more of a hard disk, a flash memory device, a dynamic random access memory (DRAM), a static random access memory (SRAM), or any combinations thereof.
• Raw/Unprocessed Data
• Raw/Unprocessed Data100 is a data that is in a raw/unprocessed fond. For example, a web page about a restaurant might say somewhere in the content of the webpage “has valet parking.” In this case, the raw data is a copy of the entire web page. In the disclosed system, an input of {“valet_parking”:true} could, for example, originate from a webpage that said “has valet parking.” As an additional example, the system may contain a data store of restaurants, for example, a data storage having restaurant-related data and/or a database having restaurant-related data. Examples of unprocessed data can include:
• • The Internet home pages of restaurants in the data store
  • Reviews of the restaurants that appear in on-line blogs
  • Reviews of the restaurants provided by individuals hired to provide data for the system
  • On-line articles about restaurants in the data store
• In the disclosed system, Raw/Unprocessed Data100 can be maintained for reprocessing (e.g., the raw data may be stored at periodic intervals so that it can be used for new runs of the batch processing system). This is advantageous because new rules, which may be developed at a later time, may be able to extract additional inputs when the raw data is reprocessed. For example, a website may have text saying “the valet scratched my car while parking it.” Even if an earlier run that evaluated content on the website failed to form any inputs about valet parking, a subsequent run may extract an input of “valet_parking”:true. Because the disclosed system can store theraw data100 for reprocessing, a batch process of the disclosed system can be rerun againstraw data100, and a new rule that understands the more complex statement could, for example, extract an input of “valet_parking”:true on a subsequent run.
• Unprocessed (Raw) Inputs
• Unprocessed Inputs350 represent the original attribute values for an entity as they were received by the disclosed system from a contributor, third party system, web page, and/or any other suitable source of information. For example, if a webpage stated somewhere in the content of the webpage “has valet parking,” the statement “has valet parking” is an Unprocessed Input. Likewise, a website about a clothing store (raw data) might contain the statement “50 percent off sale” (raw input). As another example, a contribution, from a contributor, updating the address of a business may contain “1801 ave of stars, los angeles”. Initially, the rules available when the data was first provided may have caused this input to be ignored because the address is insufficient. However, a subsequent build with improved rules could refine it to be {“address”:“1801 Avenue of the Stars”, “city”:“Los Angeles”,“state”:“CA”,“zipcode”:“90067”}.
• Unprocessed Inputs350 may, for example, be stored in one or more of the following: a file system, including a distributed file system, a data store, such as a relational or non-relational (i.e. nosql) database.
• FullProcessed Inputs and Summaries
• FullProcessed Inputs360 are Inputs that have been Processed in the most recent Batch Data Build. For example, if the raw input “has valet parking” were contained in an online restaurant review, the Batch Data Build of the disclosed system could extract an processed input of “valet_parking”: true.
• In some embodiments, each Batch Data Build may entirely replace the previous set of Full ProcessedInputs360. For example, suppose the disclosed system had entries for just one restaurant called “Joes” and five websites provided facts about the restaurant. Two websites might state that the type of food served is “Chinese”. One website might state that it's “Cantonese”. Another two websites might say that it is “Italian”. In this example, FullProcessed Inputs could include {“id”:“1”,“name”:“Joe's”, “cuisine”:“Chinese”, “source”:“website1”}, {“id”:“1”,“name”:“Joe's”, “cuisine”:“Chinese”, “source”:“website2”}, {“id”:“1”,“name”:“Joe's”, “cuisine”:“Cantonese”, “source”:“website3”},{“id”:“1”,“name”:“Joe's”, “cuisine”:“Italian”, “source”:“website4”},{“id”:“1”, “name”:“Joe's”, “cuisine”:“Italian”, “source”:“website5”}. Based on the current rules, theFullProcessed Summary760 may have {“id”:“1”, “name”:“Joe's”, “cuisine”:“Italian”} because it trusted all contributions equally and “Italian” and “Chinese” were tied while “Cantonese” was treated as an independent cuisine. In this example, a rule could be improved to determine that “Cantonese” is a type of “Chinese” cuisine and is also more specific, resulting in aFullProcessed Summary760 of
• • {“id”:“1”, “name”:“Joe's”, “cuisine”:“Chinese>Cantonese”}
    when the Batch Data Build is run. In the disclosed system,FullProcessed Inputs360 andFullProcessed Summaries760 can change when the entire table containing them is replaced while new incremental information is written to QuickProcessed tables.
• QuickProcessed Inputs and Summaries
• At the start of a Batch Data Build,QuickProcessed Inputs150 andQuickProcessed Summaries190, which represent the newly computed Inputs and Summaries since the start of the previous Batch Data Build, may be set aside or discarded, and an empty version of each ofQuickProcessed Inputs150 andQuickProcessed Summaries190 may be allocated in the data store, such as a database. For example, a user on a mobile device might notice that “Joe's” restaurant is miscategorized as “Italian”. That user, acting as a contributor, could submit a correction through software on her mobile device that sends the data to the Public application programming interface (API) (FIG. 1, 130.) That contributor's input could look like {“id”:“1”,“cuisine”:“Chinese”}. Once that input is processed, it could be saved toQuickProcessed Inputs150 and the entry for “Joes” could be re-Summarized. In this example, the new Summary for “Joes” would then be {“id”:“1”,“name”:“Joe's”,“cuisine”:“Chinese”} and because it is different than theprevious FullProcessed Summary760, the new Summary would be saved toQuickProcessed Summaries190. In the disclosed system, when determining the latest Summary for an entity, the system can check for the latest Summary inQuickProcessed Summaries190 favoring that over theFullProcessed Summary760 which only changes in a Batch Data Build.
Batch Processing and Lifecycle
• From time to time, a Batch Data Build may be run to convert Unprocessed Data and Inputs into finished view Summary Data. The output of a Batch Data Build isFullProcessed Inputs360 andFullProcessed Summaries760.
Input Processing Block
• In some embodiments, theInput Processing module145,720 can be configured to perform one or more of the extraction process, the cleaning process, the canonicalization process, the filtering process, and the validation process, each of which is described below.
• Extraction
• The Extraction step may, for example, include a selection of a fact for an attribute based on a matching rule from structured, semi-structured, and unstructured data. For example, the disclosed system could use the fact matching rule “name:[NAME]” to extract a name. In this example, the Extraction step includes selection of the name “Mc'Donalds” in a record like: {“name”:“Mc'Donalds”} using the fact matching rule “name:[NAME].” Additionally, in the Extraction step, the system could use a pattern matching rule like “***-***-****”, where the * symbol represents a wildcard character, to select the phone number “123-456-7890” from text such as: “Tel: 123-456-7890.” As an additional example, the Extraction step could interpret raw text like “This place has no high chairs for my children” to create a fact in the form: {“kid_friendly”:false}. The disclosed system can interpret raw text to create a rule by, for example, using advanced natural language processing and parsing.
• Cleaning
• The Cleaning step comprises cleaning extracted data. Cleaning extracted data may include a process to remove undesired or bad characters or entity attributes. For example, extraction of a fact matching the rule “Phone:[PHONE_NUMBER]” might incorrectly extract incorrect information such as “Phone: call now!” or extract extra information like “Phone:123-456-7890 click here”. Cleaning can discard incorrect data or remove extra data that is not desired. For example, if “Phone: call now!” were extracted, the Cleaning step could discard the data because “Phone: call now!” is incorrect data for a phone number. Additionally, if “Phone:123-456-7890 click here” were extracted, the Cleaning step could discard “click here” because “click here” is extra data that is not part of the extracted phone number. In the disclosed system, incorrect data or extra data can be discarded or removed by, for example, using two rules, such as a fact matching rule and a pattern matching rule. For example, using the fact matching rule “Phone:[PHONE_NUMBER],” the disclosed system could extract information like “Phone:123-456-7890 click here” and using the pattern matching rule “***-***-****,” the system could determine that “click here” is extra data and remove it during the Cleaning step.
• Canonicalization
• Canonicalization refers to a rules-driven step to convert data in various formats into their preferred or canonical representation. For example, one contributor may describe a phone number as “123-456-7890” and a different contributor may submit “(123)456-7890”. Converting data into a canonical representation makes it uniform and enables better entity resolution and summarization. The disclosed system can perform canonicalization by, for example, using multiple pattern matching rules and designating another pattern for the canonical representation. For example, using the pattern matching rule “***-***-****” and “(***)***-****,” with the former designated the canonical representation, the Canonicalization step could make the inputs “123-456-7890” and “(123)456-7890” uniform by representing them both as “123-456-7890.”
• Filtering
• Filtering refers to a rules-driven step to reject data that is not necessarily incorrect, but does not meet some desired criteria. This can include rejecting inputs that don't match a particular category or have insufficient confidence. For example, a Science Fiction theme restaurant might advertise that it is “located on the planet Earth in the Milky Way galaxy.” While this statement is accurate, an embodiment of the disclosed system might, for example, not have a category for the planet and galaxy where restaurants are located, and as such, the Filtering step in this example would reject the input “located on the planet Earth in the Milky Way galaxy.” Of course, in alternate embodiments, the disclosed system could have such categories. As an additional example, in an embodiment, the disclosed system could, for example, set a threshold of 100 visits for information from a website to be considered reliable. In this example, if a website that had been visited only 15 times contained the statement “it is the best store,” the system could reject the input because it does not meet the confidence rule. In other embodiments, the disclosed system could use other rules for determining confidence.
• Validation
• Validation refers to a rules driven step to reject data based on non-conformance with certain criteria. For example, a phone number field where, after canonicalization, the phone number has fewer digits than are expected for a valid phone number (e.g. Phone: 123), it is possible to reject the attribute or the entire input based on failure to meet certain criteria.
Real-Time Summarization
• Embodiments of the disclosed system may perform a Real-time Summarization process. Referring toFIGS. 1 and 2, in embodiments of the disclosed system, the QuickSummarization process module160 receivesQuickProcessed inputs150, generated by theInput Processing module145, andFullProcessed inputs360, generated by the batch processing system.
• The QuickSummarization process module160 can be configured to aggregate and filter theQuickProcessed inputs150 andFullProcessed inputs360. For example, theQuick Summarization process160 could receiveQuickProcessed inputs150 andFullProcessed inputs360 regarding valet parking. In this example,FullProcessed inputs360 might include inputs with the value “valet_parking”:false andQuickProcessed input150 might include inputs with the “valet_parking”:true. The QuickSummarization process module160 can be configured to aggregate and filter theQuickProcessed inputs150 and theFullProcessed inputs360 to create aQuickProcessed Summary190. For example, after filtering and processing, theQuickProcessed Summary190 for an entity might be “valet_parking”:true.
• In some embodiments, the QuickSummarization process module160 can be configured to maintain and index data in theQuickProcessed Inputs150 and theFullProcessed Inputs360 in a sort order determined based, at least in part, on one or more of the entity identifier, the identifier of the contributor or a user account that provided the data, the technology used to extract the data, the source or citation for the data, and/or a timestamp. To this end, a connection or iterator is created to read data simultaneously starting from the first Input with the desired entity identifier from both the QuickProcessed Inputs and the FullProcessedInputs. In each case, the iterator is advanced on either QuickProcessed Inputs or FullProcessed Inputs, whichever has the earlier timestamp. Whenever any of the attributes enumerated above except for the timestamp change, the previous input is added to the pool being considered while the others are ignored, thus allowing the system to efficiently consider only the latest version of an input from a given user, extraction technology, and citation.
• TheDiff process module200 can be configured to compareQuickProcessed Summaries190, generated by the QuickSummarization process module160, and theFullProcessed Summaries760. For example, it might compare aQuickProcessed Summary190 with the value “valet_parking”:true and aFullProcessed Summary760 with a value “valet_parking”:false. Based on the comparison, theDiff process module200 could then broadcast the result. For example, it could broadcast that theFullProcessed Summaries770 from the previous batch build indicates that there is no valet parking, whereas theQuickProcessed Summary190 from the newly computed Inputs and Summaries since the start of the previous Batch Data Build, indicates that there is valet parking.
Real-Time Processing Workflow
• Referring toFIGS. 1 and 2, as indicated by the arrow Real-time10, the upper portion of the diagram generally depicts real-time components of the system. The system receivesExternal Contributions100 as inputs.External Contributions100 include bulk data contributions, web documents, and real-time submissions from contributors. For example, the system can receive bulk data contributions such as entire websites or data stores, web documents such as individual web pages, and real-time submissions such as reviews on a website.
• One source ofExternal Contributions100 are User Writes110. User Writes110 could include, for example, direct input from contributors on a web form or a mobile device.
• In some embodiments, the system can receive User Writes110 viaPublic API module130. For example, User Writes110 could be received through a publicly accessible endpoint such as a website that submits to thePublic API module130 or through software on a website or mobile device that submits to thePublic API module130. User Writes110 may include identifiers for the contributor, origin, and developer added to them for consideration in summarization.
• An input such as aUser Write110 may have an entity identifier (e.g., entity_id) already included with it. An entity identifier can be a string of letters and numbers. In some embodiments, an entity identifier can signify that the input is an update to an existing entity. If the input does not have an identifier, the system can determine and assign a temporary identifier, referred to as a QuickProcessed Identifier, using theResolve process module120. TheResolve process module120 can be configured to assign an entity identifier to an input or to match one representation of a record to another. This makes it possible to cluster similar inputs together and assign those that reference the same entity with a common entity identifier. In many cases, inputs have different attributes but reference the same entity. The Resolve process can be used to compare inputs, determine that the inputs reference the same entity, and assign a common entity identifier to those inputs.
• In some cases, theResolve process module120 can be configured to assign an identifier as a surrogate key generated by a) random assignment, b) concatenating one or more input attributes (e.g. name+address), c) consistent hashing of one or more input attributes (e.g. md5(name+address)), or d) taking the assigned id of an existing input if a sufficiently similar input exists (e.g. name, value, phone of new input is similar enough to name, value, phone of existing input) and generating a new surrogate key when it is not.
• Once a QuickProcessed Identifier is determined for the input,Internal API module140 can receive the input fromPublic API module130. Before it is saved to storage, a copy of the original input, in its raw form, can be made. The raw copy can be saved to storage forUnprocessed Inputs350 so that it can later be reprocessed in batch with updated software or subjected to more expensive computation, including software that does entity identifier assignment.
• Additionally,Internal API module140 can be configured to interact with theStitch process module155 for rules driven moderation. TheStitch process module155 for rules driven moderation can be configured to display or provide data submissions that match certain criteria to a human moderator or more expensive machine processes for further evaluation. For example, a new restaurant owner might wish to drive business to his restaurant by diverting it from nearby restaurants. That restaurant owner might sign up for an account as a contributor of one of the disclosed system's customers and submit information that all of the other restaurants are closed. The system could then determine that a new contributor who has never interacted with the system has, on one day, reported that several local businesses have closed, causing a rule in the system that looks for such patterns to flag those submissions and enqueue them for review by a human moderator. The human moderator could in turn determine that the businesses are indeed still open and reject the submissions and further blacklist the contributor such that additional submissions will be ignored.
• At the same time, the original raw input can be processed through software that performs Extraction, Cleaning, Canonicalization, and Validation, as described above, producing aQuickProcessed Input150. In some cases, theQuickProcessed Input150 may have a QuickProcessed Identifier attached. If the QuickProcessed Input passes Validation, it can be saved to storage forQuickProcessed Inputs150 and it can move forward in the process toQuickSummarization process module160.
• Real-timeQuickSummarization process module160 can be configured to analyze and combine theQuickProcessed Inputs150 andFullProcessed Inputs360 for an entity in substantially real-time.QuickProcessed Inputs150 can represent new real-time inputs since the last Batch Data Build process.FullProccessed Inputs360 are inputs generated from a previous Batch Process by the batch processing system. Together, they comprise the full set of Inputs for each entity. For example, the Real-timeQuickSummarization process module160 could receiveQuickProcessed inputs150 andFullProcessed inputs360 regarding valet parking. In this example,FullProcessed inputs150 might include inputs with the value “valet_parking”:false andQuickProcessed input360 might include inputs with the “valet_parking”:true. TheQuickSummarization process module160 could then aggregate theQuickProcessed inputs150 andFullProcessed inputs360 and then filter them usingHigh Confidence Filter170 andLow Confidence Filter180 to create aQuickProcessed Summary190. For example, after filtering and processing, theQuickProcessed Summary190 for an entity might be “valet_parking”:true.
• QuickProcessed Inputs and FullProcessed Inputs may be stored in adata store150,360. Thedata store150,360 can be respectively clustered by an entity identifier (e.g. a uuid such as 0e3a7515-44e0-42b6-b736-657b126313b5). This can allow re-Summarization to take place quickly as new Inputs are received such that Inputs only pertaining to an entity for which a new Input has been received are processed.QuickProcessed Inputs150 andFullProcessed Inputs360 may be stored in a sort order such that it facilitates processing the data in streams or skipping over Inputs that are determined to be superseded by Inputs that are, for example, newer submissions from the same submitter or citing an identical reference.
• In some embodiments, theQuickSummarization process module160 can read Inputs concurrently from the QuickProcessedInput data store150 and FullProcessedInput data store360, to facilitate choosing only the Inputs that need to be considered in Summarization. The summarization of the Inputs can be represented or displayed using a view (e.g., a materialized view). A view is one possible summarization of the Inputs and representation of entities according to one or more rules. The one or more rules can determine which entities are included, which attributes are included for each entity, what indexing optimizations are performed, and what additional attributes and attribute variations are computed for each entity. In some embodiments, a view is uniquely identified by a view_id. Data stores often track views in system tables and these system tables contain metadata about views. In our case, a view is assigned an identifier and that identifier is used to lookup metadata about the view from the data stores, such as the names of the attributes, their datatype, sort preferences, indexing rules, etc.
• For each view associated with the dataset to which the Input was assigned, a QuickSummarization Process can be performed. Views may have different rules about attributes to be computed, the rules that apply to those attributes,confidence thresholds170,180 for the Summary entity, and other software rules and transformations. Each QuickSummarization process can produce a QuickProcessed Summary for each View. Each QuickProcessed Summary is compared to the most recent Summary retrieved from the QuickProcessedSummary data store190 or the FullProcessedSummary data store760. If the QuickProcessed Summary is different than the previous version, the new QuickProcessed Summary is saved to the QuickProcessedSummary data store190 and a Diff record is produced. A Diff record can include a row of data that includes, for example, (1) an entity identifier of the entity whose attributes have changed and (2) the changed attributes. The Diff record may include an entire copy of the new Summary or the attributes that are different from the previous Summary. The Diff record is saved to a Diff data store and published over the network to processes that listen for Diff records and update Materializations of Summary data.
• The following is an example of a Diff record in one possible embodiment: “timestamp”:1363321439041, “payload”: {“region”:“TX”, “geocode_level”:“front_door”, “tel”:“(281) 431-7441”, “placerank”:90,“category_labels”:[[“Retail”,“Nurseries and Garden Centers”]], “searchtags”:[“Houston”,“Grass”,“South”],“name”. “Houston Grass”,“longitude”:“-95.464476”, “fax”:“(281) 431-8178”, “website”:“http://houstonturfgrass.com”, “postcode”:“77583”, “country”:“us”, “category_ids”:[164],“category”:“Shopping >Nurseries & Garden Centers”,“address”:“213 McKeever Rd”, “locality”:“Rosharon”, “latitude”:“29.507771”}, “type”:“update”,“factual_id”:“399895e6-0879-4ed8-ba25-98fc3e0c983f”, “changed”: [“address”,“tel”]}. In this example, the Diff record indicates that the address and telephone for Houston Grass have changed, which can result in an update to each copy of the materialized data store or index rows for that entity.
• Embodiments of the disclosed system may include materialized data stores orindexes510,520. Materialized data stores orindexes510,520 are searchable relational or non-relational data stores or search index servers. In some embodiments, the materialized data stores orindexes510,520 can be associated with a particular application domain or a particular data service. The disclosed system may utilize data store systems like Postgre SQL (relational) and Apache Solr (non-relational, search server) interchangeably, sometimes for the same data, and can choose the one that best services the type of query requested. For example, the disclosed system can receive a query for data associated with a particular view or type of data. In response, the disclosed system can determine one or more of a type of the query, a type of the entity, an application or a device that sends the query, an application domain associated with the entity, or any relevant information associated with the query to determine one of the data store systems or a combination of such systems to use to respond to the query. Then the disclosed system can use the determined one of the data store systems or combinations of systems to respond to the query.
Batch Processing Workflow
• Referring toFIGS. 1 and 2, as indicated by thearrow Batch20, the lower portion of the diagram generally illustrates batch processing components of the system.
• The Batch Processing Workflow can receive Large Uploads andBulk Contributions700 such as
• Raw Inputs700
• Universally Unique Identifier (uuid) attachment data
• Message Digest 5 (md5) attachment data
• In addition, the Batch Processing Workflow uses previously processed data such as previous FullProcessedInputs and previous QuickProcessedInputs for steps such as UUID Retention and Diff Generation. These steps are described below.
• Pre Batch Build
• Prior to initiating the Batch Build process, the real-time processed data can be provided to adata store710, such as a Hadoop Distributed File System (HDFS), so they can be used as inputs for the Batch Build. When this step is initiated, the time can be recorded, that can be used during the Catchup Phase.
• Specifically, the following data can be provided to the data store710:
• QuickProcessedSummaries190—summaries that have been created since the last Batch Data Build. This may include brand new summaries, deleted summaries and summaries that have certain fields updated
• Unprocessed Inputs350—the raw inputs that have been written to this dataset since the last Batch Data Build
• new uuid mappings—a mapping of input ids to entity ids for new summaries that were generated since the last batch run
• In some embodiments, quick processed inputs from the previous version of the data are not used, except for UUID Retention (described below). In such embodiments, this is accounted for by using theUnprocessed Inputs350 instead. This ensures that inputs are completely reprocessed.
Batch Build
• The Batch Build is a process in which the data can be processed and made ready for loading into production.
• Input Processing:
• Raw inputs700 andUnprocessed Inputs350 are fed into theInput Processing module720 fromHDFS710. The Extract step may not preserve any notion that the data was previously extracted. Extraction may be done on theraw inputs700.
• The Extract step may use a rule framework that canonicalizes, cleans, fills in values and filters inputs as described above and as illustrated in the examples below.
• 123 main street=>123 Main St.
• city: Los Angeles=>city: Los Angeles, state: CA
• The Extract step can also sort out which inputs should be attached and which inputs should be batch-resolved. The Extract step can optionally determine that some inputs should be reviewed by a human, a computationally powerful process, or a third party API. The Extract step can set a moderation action flag within the metadata of the input and insert it directly or via API into the Stitch data store, which is used to coordinate relatively costly processes such as moderation.
• Batch Resolve:
• Batch resolve, which can be performed by theResolve process module722, may take the extracted inputs and group them based on whether they represent the same entity or not and assigns a unique id to each set of inputs. For example, batch resolve can assign a unique id generated by a) random assignment, b) concatenating one or more input values from a set of inputs (e.g. name+address), c) hashing of one or more values form an asset of inputs (e.g. md5(name+address)), or d) taking the assigned id of an existing set of inputs if a sufficiently similar input exists (e.g. name, value, phone of new input is similar enough to name, value, phone of existing input).
• UUID Retention:
• After thebatch resolve module722 completes its process, theUUID Retention module725 can be initiated. An objective of theUUID Retention module725 can include modifying the identifier associated with entities (e.g., entity_id) so that a single entity, such as the Eiffel Tower, can maintain the same entity identifier even when input data is re-processed by the batch processes (e.g., across multiple batch runs). This enables an entity to be associated with the same identifier even when data associated with the entity is re-processed multiple times.
• This is accomplished, for example, by reading in theprevious FullProcessedInputs360, and generating a mapping file or table, which includes a mapping between an input_id an entity_id. An input_id is a unique identifier assigned to each set of attributes coming from a single input data contribution. For example, all of the attributes pulled from the homepage of the French Laundry, such as the name, address, phone number constitute one input data contribution. Lots of other websites and contributors can also provide input data contributions describing the French Laundry. Each of these input data contributions has its own input_id that uniquely identifies it from other input data contributions. The input_id can include a message digest 5 (md5) hash of an input data contribution. In contrast, an entity_id is an identifier assigned to all of the input data contributions and the summary record of the French Laundry (currently a UUID).
• In some embodiments, the input_id to entity_id mapping can be combined with the mappings for newly written summaries. For example, each input in FullProcessed Inputs and QuickProcessed Inputs has an input_id that uniquely identifies the original unprocessed input and an entity_id representing the entity associated with the FullProcessed Inputs and QuickProcessed Inputs, as determined by theResolve process module120. Using the input_id to entity_id mapping, the input_id can be used to assign the original entity_id to all inputs data items that are within the same set.
• In some embodiments, an example of UUID Retention could be as follows:
• Mapping:
• • input_id_0, original_entity_id
• Input Set:
• • input_id_0, new_entity_id, data
  • input_id_1, new_entity_id, data
  • input_id_2, new_entity_id, data
• In this example, in the previous batch build, input_id_0, had the entity id “original_entity_id”. In the current Batch Build, since the sample input set contains the input with input id: input_id_0, the disclosed system can map all inputs in the input set to original_entity_id. As such, in this example, the end result would be as follows
• End Result:
• • input_id_0, original_entity_id, data
  • input_id_1, original_entity_id, data
  • input_id_2, original_entity_id, data
• The output of theUUID Retention module725 can include a grouped set of inputs that have the same entity_ids as they had in the previous batch run, as well as preserving any entity_ids that were generated in between batch runs. As described above, theUUID Retention module725 can preserve the same UUID for the same entity across Batch Builds.
• In some cases entities may be merged or split, depending on the result of Batch Resolve. TheUUID Retention module725 can, in effect, specify how to deal with split and merge cases. For example, in the case of a merge, it may be preferred to use the entity_id with the greater number of inputs. In the case of a split, the entity_id may be assigned to the input set that has the greater number of inputs and generate a new id for the input cluster forming the new summary. This behavior can be customized depending on the dataset and desired outcome.
• Data Attachment
• After UUID Retention, the Data Attachment process may be performed by theData Attachment module727. A purpose of the Data Attachment process can be to attach inputs that are (1) unresolvable, (2) derived from a summary, (3) derived from an input, or (4) for inputs that with a sufficient degree of confidence, pertain to a specific entity_id, such as contributor edits to a specific entity_id or an input that has geocode information pertaining to a specific input.
• Data attachment can be based on an entity_id or an input_id. For example, theData Attachment module727 can be configured to attach (or combine) a source input to a set of inputs generated from UUID Retention when the source input has the same entity_id as that of the set of inputs. As another example, theData Attachment module727 can be configured to attach (or combine) a source input when the source input is associated with the same parent input_id as that of the set of inputs, where the parent input id refers to a unique identifier of the input to which the source input should be attached. These examples are illustrated with the following embodiments.
• In some embodiments, an example of Entity_ID Data Attachment could be as follows:
• Attachment Data:
• • input_id_0, entity_id_0, data
• Input Set:
• • input_id_1, entity_id_0, data
  • input_id_2, entity_id_0, data
  • input_id_3, entity_id_0, data
• In this example, since the sample input set and the source data have the same entity_id: entity_id_0, the attachment data is added to the sample input set.
• End Result:
• • input_id_0, entity_id_0, data
  • input_id_1, entity_id_0, data
  • input_id_2, entity_id_0, data
  • input_id_3, entity_id_0, data
• In some embodiments, an example of Input_ID Data Attachment could be as follows:
• Attachment Data:
• • input_id_0, (no entity id), parent_input_id: input_id_1, data
• Input Set:
• • input_id_1, entity_id_0, data
  • input_id_2, entity_id_0, data
  • input_id_3, entity_id_0, data
• In this example, since the sample input set and contains an input that matches the parent input id of the source data, the attachment data is added to the sample input set.
• End Result:
• • input_id_0, entity_id_0, data
  • input_id_1, entity_id_0, data
  • input_id_2, entity_id_0, data
  • input_id_3, entity_id_0, data
• Extended Attribute Set Extraction
• Extended Attribute Set Extraction is an additional extraction process performed by theextended attributes module728. Theextended attributes module728 can be configured to run extraction on certain inputs to extract an “extended attribute set”. An extended attribute set may not be a part of the core attribute set, but may contain information that pertains to specific views. For example, “vegan” is an attribute that would pertain to a restaurant view but not to a doctor's view.
• In some embodiments, rules may be written in a rules framework that determines whether a set of inputs is re-extracted for extended attributes. For example, if a set of inputs has a single input that has the category “Restaurant”, all inputs in that input set can be re-extracted for extended attributes pertaining to restaurants.
• The output of theextended attributes module728 includefinal inputs729. The final inputs can be stored in theFullProcessed Inputs360 storage, which may relay the final inputs to thequick summarization module160.
• Summarization
• Summarization module730 is configured to perform a summarization process. The summarization process includes a process by which the final representation of a set of inputs representing the same entity can be generated.Summarization module730 can use a rules framework to generate a summary based on thefinal inputs729. Each dataset may have multiple views, including side-effect views. Each set of inputs may generate multiple view summaries. Each of the summaries generated from the same set of inputs have the same entity_id.
• A side-effect view includes a new view (e.g., set of summary entities) that does not have a one-to-one relationship with the entity id for the given inputs. A side-effect view can be generated as a by-product of other views and their inputs rather than directly producing summaries from the associated entity inputs. The side-effect view allows thesummarization module730 to provide an arbitrary number of summary records (e.g., an arbitrary number of related entities) from a single data input. One such example of this is Crosswalk, which is a view that links entity_ids to specific input sources. For instance, the side effect view creation process can determine whether an input data matches a rule, such as “is a namespace we track in crowsswalk” (e.g. because it has a url like webname.com/[some_place_id]), and once there is a match, the side effect view creation process can create a new entity, for example, with {“namespace”:“webname”, “id”:“[some_place_id]”, “factual_id”:“[id_of_referenced_entity]”}. Therefore, even if the input data is already associated with an entity, the side effect view creation process can generate additional entities associated with the input data based on a rule maintained by the side effect view creation process.
• FollowingSummarization730, the results may be filtered withHigh Confidence filter740 andLow Confidence Filter750 as described previously, and stored asFullProcessed Summaries760.
• Data store Format Generation process
• In this process, theFullProcessedInputs360 andFullProcessedSummaries760 are built.FullProcessedInputs360 can include all inputs for a given dataset and can be organized in a way where entity_id lookup and summarization is efficient.FullProcessedSummaries760 can contain all the summary records for all views in a given dataset, organized in a way where entity_id and view_id lookup is efficient. These files can be bulk loaded into a data store during a MakeLive step. The output of these this step is represented by729,740, and750 inFIG. 1.
• Diff Generation
• Diff Generation module770 can be configured to generate all the “diff” records that comprise the difference between the current batch run and the prior real-time updated dataset and output them to Diff API to DownloadPartners500, which allows authorized partners to download the difference records from the system. Each such record can be referred to as a “diff.” Specific diff types are described above. Diffs can be generated by comparing each summary for a view against the prior version of the summary for that same view. Diffs can be generated for every view for each summary. The current summaries can be compared against theprior FullProcessedSummaries760 andprior QuickProcessedSummaries190 tables. The same diff generation mechanism can be used to generate the diffs for theindexes510,520, and the diff for third parties to be provided via thediff API500.
• The Diffs are also written to the Data store Format, which allows for efficient lookup based on date and entity_id.
• Materialization Build
• Materialization Build module780 is configured to produce an output format that is ready for serving other computing systems, such as data stores. For example, theMaterialization Build module780 can be configured to build an inverted index (e.g., a data store) that allows for searching of the inputs. In some embodiments, theMaterialization Build module780 can be configured to build a materialization on a per-view basis. In other embodiments, theMaterialization Build module780 can be configured to build a materialization that includes multiple views.
• In some embodiments, a simplified example of an inverted index materialization can include the following:
• Sample Data:
• • doc_id_0, entity_id_0, view_id, Business, San Diego, Calif.
  • doc_id_1, entity_id_1, view_id, Business, San Francisco, Calif.
• Index:
• • entity_id_0: {doc_id_0}
  • entity_id_1: {doc_id_1}
  • Business: {doc_id_0, doc_id_1}
  • San: {doc_id_0, doc}
  • Diego: {doc_id_0}
  • Francisco: {doc_id_1}
  • CA: {doc_id_0, doc_id_1}
• Using the simplified index in the example, the data can be easily searchable by keyword or other attributes. For example, searching for “Diego”, would yield summaries for doc_id_0 and doc_id_1 in this example.
• If there are multiple views per materialization, the view_id could be used as an additional keyword filter for searches.
• In some embodiments, each materialized data store can be associated with a particular application domain, a particular service, or a particular view. Therefore, when a system receives a query for data, the system can determine, based on the particular application domain, the particular service, and/or the particular view associated with the query and/or requested data, one or more of the materialized data stores to serve the query.
Batch Processing MakeLive
• MakeLive is a process by which a Batch Build can be put into production. The MakeLive process can be accomplished through Data store Loading, Catchup and New
• Materialization Notification. After the MakeLive process is completed, all API requests can use the newly batch-built data.
• Data Store Loading
• Once a Batch Data Build passes all required regression and other Quality Assurance tests, a new table in the data store can be created with a new version number forFullProcessedInputs360,FullProcessedSummaries760,QuickProcesesdInputs150, andQuickProcessedSummaries190. The data store format files (FullProcessedInputs360, FullProcessedSummaries760) can be loaded into their respective new tables.Diffs200/770 can be appended to an existing DiffTable.
• InFIG. 1, the Real-Time Processing can refer to newly builtFullProcessedInputs360 andFullProcessedSummaries760 through a data store-api-server once data store loading is complete. This can be accomplished by changing the pointer of theFullProcessedInputs360 andFullProcessedSummaries760 tables so that the newer tables are visible to Real-Time Processing and the older references are no longer visible to Real-Time Processing.
• An example of the loading ofFullProcessedInputs360 is illustrated in the transition from729 to360 inFIG. 1. An example of the loading ofFullProcessedSummaries760 is illustrated in the transition from740,750 to760 inFIG. 1.
• Catchup Phase
• During the time between when the batch run was started and when the Catchup Phase is first initiated, the data store may have taken additional real-time writes that were not processed during our Batch Build step. The real-time writes can refer to any writes that have been received in real-time and have generated QuickProcessed inputs. Once the Batch Build step is completed, thereby creating a newly batch built dataset, the Catchup Phase may update the newly batch built dataset, maintained in the indexeddata stores510,520, or a Diff API to Downloadpartners500, based on these new real-time writes, so that the newly batch built dataset becomes up to date with the additional real-time writes.
• FIG. 3 illustrates the Catchup process in accordance with some embodiments. To accomplish Catchup, the Quick ProcessedInputs810 from the prior version of the dataset can be each copied into the new Quick ProcessedInputs820, based on whether the timestamps of those inputs are after the timestamp at which the batch run was initiated. Specifically, each input in Quick ProcessedInputs810 can be added to the new Quick ProcessedInputs820 with the same entity_id (if it exists) for the new QuickProcessedInput table. If the same entity_id doesn't exist in the new QuickProcessedInput table, it can create a brand new input set for that entity_id. For each entity_id with an additional input,re-summarization830 can be performed for all views. If the generated summaries are different from the inputs from FullProcessedSummaries, a diff is written to theDiffTable840. Thematerialization880 is in turn updated by anynew Diffs840.
• New Materialization Data Store Notification
• The final step for making a batch built dataset into production-ready dataset can include a process for enabling the FullProcessedInputs, FullProcessedSummaries, QuickProcessedInputs, and QuickProcessedSummaries tables. A flag can be cleared and the Summary Materialization versions can be updated to point to the newly built ones. This process can change the pointer from previous versions of510 and520 with the newest versions of the materializations built by the latest Batch Build.
• The Unprocessed Inputs provided by the Real-Time Workflow at the Pre Batch Build step can be copied into theUnprocessed Inputs350, so that they can be processed by the next Batch Data Build. TheUnprocessed Inputs350 can be deduplicated to prevent duplicate entries.
• After these steps, all updates to the data can be handled by the Real-Time Data Processing workflow, until the next Scheduled Batch Data Build.
• Embodiments of the disclosed system can be used in a variety of applications. For example, embodiments of the disclosed system can be used to gather and summarize data from various application domains, such as social networking, online advertisements, search engines, medical services, media services, consumer package goods, video games, support groups, or any other application domains from which a large amount of data is generated and maintained.
Executable Code Embodiments
• Embodiments of the disclosed system may be built upon logic or modules comprising executable code. The executable code can be stored on one or more memory devices. Accordingly, a logic does not have to be located on a particular device. In addition, a logic or a module can be multiple executable codes located on one or more devices in the systems disclosed herein. For instance, access logic responsive to an input for accessing and retrieving data stored in one or more cells in the data store can be one executable code on an application server. In alternative embodiments, such access logic is found on one or more application servers. In still other embodiments, such access logic is found on one or more application servers and other devices in the system, including, but not limited to, “gateway” summary data servers and back-end data servers. The other logics disclosed herein also can be one or more executable code located on one or more devices within a collaborative data system.
• In certain embodiments, the disclosed systems comprise one or more application servers, as well as one or more summary data servers, and one or more back-end data servers. The servers comprise memory to store the logics disclosed herein. In particular embodiments, the one or more application servers store the logics necessary to perform the tasks disclosed herein. In other embodiments, the summary servers store the logics necessary to perform the tasks disclosed herein. In other embodiments, the back-end servers store the logics necessary to perform the tasks disclosed herein.
• In certain embodiments, the client web browser makes requests to the one or more application servers. Alternatively, the disclosed systems comprise one or more summary or back-end data servers to which the client web browser makes requests.
• In an exemplary embodiment, the one or more application servers receive requests from the client web browser for specific data or tables. Upon these requests, the one or more application servers calls upon one or more data store servers to request summary or detail data from cells or tables. The one or more application servers also call upon the one or more data store servers when a request to submit new data inputs is made. The one or more application servers receive the data from the one or more summary servers and the one or more application servers generate HTML and JavaScript objects to pass back to the client web browser. Alternatively, the one or more application servers generate XML or JSON to pass objects through an API.
• In one embodiment, the data store servers are based on an architecture involving a cluster of summary data servers and a cluster of back-end data servers. Note, however, that a system could include a single summary server and back-end data server. In this embodiment, the array of summary data servers are utilized to request from back-end data servers, summary data and attributes of such summarized data points (confidence, counts, etc.). The array of summary servers also caches such summary data and summary attributes so that faster access to such summary data can be access without the need for an additional request to the back-end data server.
• The present systems and processes rely on executable code (i.e., logic) stored on memory devices. Memory devices capable of storing logic are known in the art. Memory devices include storage media such as computer hard disks, redundant array of inexpensive disks (“RAID”), random access memory (“RAM”), and optical disk drives. Examples of generic memory devices are well known in the art (e.g., U.S. Pat. No. 7,552,368, describing conventional semiconductor memory devices and such disclosure being herein incorporated by reference).
• Other embodiments are within the scope and spirit of the disclosed subject matter.
• The subject matter described herein can be implemented in digital electronic circuitry, or in computer software, firmware, or hardware, including the structural means disclosed in this specification and structural equivalents thereof, or in combinations of them. The subject matter described herein can be implemented as one or more computer program products, such as one or more computer programs tangibly embodied in an information carrier (e.g., in a machine-readable storage device), or embodied in a propagated signal, for execution by, or to control the operation of, data processing apparatus (e.g., a programmable processor, a computer, or multiple computers). A computer program (also known as a program, software, software application, or code) can be written in any form of programming language, including compiled or interpreted languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program does not necessarily correspond to a file. A program can be stored in a portion of a file that holds other programs or data, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers at one site or distributed across multiple sites and interconnected by a communication network.
• The processes and logic flows described in this specification, including the method steps of the subject matter described herein, can be performed by one or more programmable processors executing one or more computer programs to perform functions of the subject matter described herein by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus of the subject matter described herein can be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).
• Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processor of any kind of digital computer. Generally, a processor will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a processor for executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. Information carriers suitable for embodying computer program instructions and data include all forms of non-volatile memory, including by way of example semiconductor memory devices, (e.g., EPROM, EEPROM, and flash memory devices); magnetic disks, (e.g., internal hard disks or removable disks); magneto-optical disks; and optical disks (e.g., CD and DVD disks). The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
• To provide for interaction with a user, the subject matter described herein can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, (e.g., a mouse or a trackball), by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well. For example, feedback provided to the user can be any form of sensory feedback, (e.g., visual feedback, auditory feedback, or tactile feedback), and input from the user can be received in any form, including acoustic, speech, or tactile input.
• The techniques described herein can be implemented using one or more modules. As used herein, the term “module” refers to computing software, firmware, hardware, and/or various combinations thereof. At a minimum, however, modules are not to be interpreted as software that is not implemented on hardware, firmware, or recorded on a non-transitory processor readable recordable storage medium. Indeed “module” is to be interpreted to include at least some physical, non-transitory hardware such as a part of a processor or computer. Two different modules can share the same physical hardware (e.g., two different modules can use the same processor and network interface). The modules described herein can be combined, integrated, separated, and/or duplicated to support various applications. Also, a function described herein as being performed at a particular module can be performed at one or more other modules and/or by one or more other devices instead of or in addition to the function performed at the particular module. Further, the modules can be implemented across multiple devices and/or other components local or remote to one another. Additionally, the modules can be moved from one device and added to another device, and/or can be included in both devices.
• The subject matter described herein can be implemented in a computing system that includes a back-end component (e.g., a data server), a middleware component (e.g., an application server), or a front-end component (e.g., a client computer having a graphical user interface or a web browser through which a user can interact with an implementation of the subject matter described herein), or any combination of such back-end, middleware, and front-end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), e.g., the Internet.
• The terms “a” or “an,” as used herein throughout the present application, can be defined as one or more than one. Also, the use of introductory phrases such as “at least one” and “one or more” should not be construed to imply that the introduction of another element by the indefinite articles “a” or “an” limits the corresponding element to only one such element. The same holds true for the use of definite articles.
• It is to be understood that the disclosed subject matter is not limited in its application to the details of construction and to the arrangements of the components set forth in the following description or illustrated in the drawings. The disclosed subject matter is capable of other embodiments and of being practiced and carried out in various ways. Also, it is to be understood that the phraseology and terminology employed herein are for the purpose of description and should not be regarded as limiting.
• As such, those skilled in the art will appreciate that the conception, upon which this disclosure is based, may readily be utilized as a basis for the designing of other structures, methods, and systems for carrying out the several purposes of the disclosed subject matter. It is important, therefore, that the claims be regarded as including such equivalent constructions insofar as they do not depart from the spirit and scope of the disclosed subject matter.
• Although the disclosed subject matter has been described and illustrated in the foregoing exemplary embodiments, it is understood that the present disclosure has been made only by way of example, and that numerous changes in the details of implementation of the disclosed subject matter may be made without departing from the spirit and scope of the disclosed subject matter.

Claims (22)

We claim:

1.-23. (canceled)

24. A method comprising:

generating a first summary data using a set of data, the first summary data includes a first entity identifier and a first value associated with the first entity identifier;

generating a second summary data using the first set of data and a second set of data, the second summary data includes a second entity identifier and a second value associated with the second entity identifier;

determining a difference between the first summary data and the second summary data; and

updating the first summary data based upon the difference between the first summary data and the second summary data.

25. The method ofclaim 24, wherein the first set of data comprises bulk data input.

26. The method ofclaim 25, wherein the bulk data input comprises one or more of:

raw information received from one or more contributors;

web-crawler data received from a web-crawler; or

data received from a storage center.

27. The method ofclaim 25, wherein the second set of data comprises intermittent data.

28. The method ofclaim 27, wherein the intermittent data comprises real-time data submissions.

29. The method ofclaim 27, further comprising:

formatting the bulk data input into structured data;

group a plurality of elements in the structured data; and

generate an entity identifier for the plurality of elements.

30. The method ofclaim 24, wherein the first summary data comprises a first entity identifier and the second summary data comprises a second entity identifier, and wherein determining the difference between the first summary data and the second summary data comprises:

determining whether a first value of the first entity identifier and a second value of the second entity identifier are equal; and

when the first value and second value are equal, comparing data associated with the first entity identifier and the second entity identifier.

31. A non-transitory computer-readable storage medium comprising computer-executable instructions that, when executed by at least one processor, cause the at least one processor to perform a method comprising:

generating a first summary data using a set of data, the first summary data includes a first entity identifier and a first value associated with the first entity identifier;

generating a second summary data using the first set of data and a second set of data, the second summary data includes a second entity identifier and a second value associated with the second entity identifier;

determining a difference between the first summary data and the second summary data; and

updating the first summary data based upon the difference between the first summary data and the second summary data.

32. The non-transitory computer-readable storage medium ofclaim 31, wherein the method further comprises:

generate a third data set by combining the first data set and the second data set; and

generate third summary data for the third data set.

33. The non-transitory computer-readable storage medium ofclaim 31, wherein the first data set comprises one or more of:

raw information received from one or more contributors;

web-crawler data received from a web-crawler; or

data received from a storage center.

34. The non-transitory computer-readable storage medium ofclaim 33, wherein the second data set comprises data received from a user to correct information in the first data set.

35. The non-transitory computer-readable storage medium ofclaim 34, wherein the first summary data comprises a first entity identifier and the second summary data comprises a second entity identifier, and wherein determining the difference between the first summary data and the second summary data comprises:

determining whether a first value of the first entity identifier and a second value of the second entity identifier are equal; and

when the first value and second value are equal, comparing data associated with the first entity identifier and the second entity identifier.

36. The non-transitory computer-readable storage medium ofclaim 31, wherein the method further comprises processing the first data set to generate a first structured data set.

37. The non-transitory computer-readable storage medium ofclaim 36, wherein the second data set comprises real-time data submissions, and wherein the method further comprises processing the second data set to generate a second structured data set in response to receiving the real-time data submissions.

38. A system comprising:

at least one processor; and

memory encoding computer-executable instructions that, when executed by the at least one processor, perform a method comprising:

generating a first summary data using a set of data, the first summary data includes a first entity identifier and a first value associated with the first entity identifier;

generating a second summary data using the first set of data and a second set of data, the second summary data includes a second entity identifier and a second value associated with the second entity identifier;

determining a difference between the first summary data and the second summary data; and

updating the first summary data based upon the difference between the first summary data and the second summary data.

39. The system ofclaim 38, wherein the first set of data comprises bulk data input.

40. The system ofclaim 39, wherein the bulk data input comprises one or more of:

raw information received from one or more contributors;

web-crawler data received from a web-crawler; or

data received from a storage center.

41. The system ofclaim 39, wherein the second set of data comprises intermittent data.

42. The system ofclaim 41, wherein the intermittent data comprises real-time data submissions.

43. The system ofclaim 41, further comprising:

formatting the bulk data input into structured data;

group a plurality of elements in the structured data; and

generate an entity identifier for the plurality of elements.

44. The system ofclaim 24, wherein the first summary data comprises a first entity identifier and the second summary data comprises a second entity identifier, and wherein determining the difference between the first summary data and the second summary data comprises:

determining whether a first value of the first entity identifier and a second value of the second entity identifier are equal; and

when the first value and second value are equal, comparing data associated with the first entity identifier and the second entity identifier.

Images