Disclosure of Invention
The present invention has been made in view of the above-described problems occurring in the prior art.
The invention provides a commodity recommendation method based on big data analysis, which solves the problems of insufficient recommendation precision and individuation degree caused by single integration of multi-source heterogeneous data and mining of interactive relations in the prior art.
In order to solve the technical problems, the invention provides the following technical scheme:
in a first aspect, the present invention provides a commodity recommendation method based on big data analysis, which includes:
Collecting the public opinion data of commodities from a social platform and a comment area, integrating the behavior data, the store data and the public opinion data to form a recommendation data set and preprocessing the recommendation data set;
Extracting behavior characteristics from behavior data, extracting store characteristics from store data, and extracting public opinion characteristics from public opinion data;
integrating the behavior features, store features and public opinion features into comprehensive feature vectors, and generating user portraits according to the comprehensive feature vectors;
Based on the user portraits, constructing a commodity recommendation model by utilizing a machine learning algorithm, taking the user portraits as the input of the model, and outputting a commodity recommendation list by combining public opinion heat of commodities and store activity data in an area by using the commodity recommendation model;
and collecting behavior data and real-time data of store activities, dynamically adjusting a recommendation list, and recording feedback of a user on recommended commodities.
The commodity recommendation method based on big data analysis is used for collecting user behavior data of a user on a platform and store data of an area where the user is located, and collecting public opinion data of commodities from a social platform and a comment area, wherein the method comprises the following specific steps of:
combining the IP address of the user equipment with GPS positioning data to preliminarily acquire the geographic position of the user;
binding the user identification and the geographic position, and initially establishing a record;
acquiring behavior data of a user on a platform through a commodity recommendation platform;
according to the geographic position of the user, the store in the area where the user is selected from a store database;
public opinion data related to the merchandise is collected from the social platform and the comment area.
The invention relates to a commodity recommendation method based on big data analysis, which is characterized in that behavior data, store data and public opinion data are integrated to form a recommendation data set and are preprocessed, wherein the method specifically comprises the following steps:
The behavior data, store data and public opinion data are integrated to form a recommended data set, and the expression is:
;
Wherein,Represent the firstA recommended data set for an individual user,Represent the firstThe behavior data of the individual user is stored,Representation and the firstStore data associated with the individual user,Representation and the firstPublic opinion data related to individual users;
Generating a unique hash value for each piece of data in the recommended data set by using a hash table, comparing whether the hash value is repeated, and deleting the repeated data after hash collision detection;
Detecting abnormal values in the recommended data set by using the quarter bit interval IQR, correcting the abnormal values, setting an abnormal threshold value, correcting data exceeding the abnormal threshold value by the average value of the similar data;
And predicting the value of missing data according to the values of K neighbor points most similar to the current data point by a K neighbor algorithm, and then carrying out missing completion on the data by a KNN interpolation method.
The invention relates to a commodity recommendation method based on big data analysis, which comprises the following specific steps of extracting behavior characteristics from behavior data, extracting store characteristics from store data and extracting public opinion characteristics from public opinion data:
Performing aggregation analysis on the behavior data, counting the behavior frequency of a user in a certain time window, extracting the distribution characteristics of behavior types, and extracting the behavior characteristics according to the formula:
;
Wherein,A feature vector representing the behavior data,The length of the time window is indicated,、、AndRespectively represent the time of the userThe number of clicks, searches, browses and purchases,Representing user behavior time period characteristics;
Counting the number of shelves and classification distribution of commodities of stores in an area where a user is located, extracting matching features of commodity categories and inventory, wherein a store feature extraction formula is as follows:
;
Wherein,A feature vector representing store data,Indicating the number of stores in the area where the user is located,Index variables representing the store's store locations,Represent the firstThe number of items on sale at a single store,Represent the firstThe preferential strength of the activities of the individual stores,Representing the user and the firstThe straight line distance of the individual store,Representing a distance weight function;
Counting the number of times of mention of the commodity in social media, extracting the heat characteristic of the commodity, analyzing comment content through an emotion analysis model, extracting the emotion tendency score of the commodity, and calculating the evaluation characteristic of the commodity by combining the heat of the commodity and the emotion tendency score, wherein a public opinion characteristic extraction formula is as follows:
;
Wherein,Feature vectors representing the public opinion data,Represents the total number of statistical public opinion data,Represent the firstThe forwarding amount of the public opinion data,Represent the firstEmotional tendency score of the strip public opinion data;
The behavior feature, the store feature and the public opinion feature are integrated into a comprehensive feature vector, and the user portrait is generated according to the comprehensive feature vector, and the specific steps are as follows:
The feature vector integration formula is:
;
Wherein,Representing the integrated comprehensive feature vector;
Based on integrated feature vectorsA user portrait for each user is constructed.
The invention relates to a commodity recommendation method based on big data analysis, which is a preferable scheme, wherein the commodity recommendation model is constructed by a machine learning algorithm based on user portraits, and comprises the following specific steps:
adopting a DeepFM model based on deep learning as a commodity recommendation model, and outputting interest prediction scores of users on commodities by combining user portraits, commodity characteristics and dynamic characteristics;
the commodity characteristics mainly refer to basic information of commodities and public opinion heat characteristics, and a public opinion heat characteristic calculation formula is as follows:
;
Wherein,Represents the public opinion popularity score of the commodity,Index variables representing the records of public opinion,Represent the firstThe forwarding amount of the public opinion data,Represent the firstThe emotion tendencies score of the strip public opinion data,A logarithmic function is represented and is used to represent,Represent the firstComment number of the strip public opinion data;
the dynamic characteristics refer to real-time update conditions of store activity data and behavior data in the area, and the store activity characteristics of the area can be calculated by the following formula:
;
Wherein,Indicating the activity intensity of the store in the area.
The invention relates to a commodity recommending method based on big data analysis, which comprises the following steps that a user image is taken as the input of a model, and a commodity recommending list is output by the commodity recommending model in combination with public opinion heat of commodities and store activity data in a region, wherein the method comprises the following specific steps:
Mapping commodity features and dynamic features to a low-dimensional dense vector space to generate feature embedded representations;
aiming at the feature intersection FM part of the commodity recommendation model, capturing the second-order interaction relation between input features by using a factorizer to obtain the output of the FM part;
Aiming at the nonlinear feature learning Deep part of the commodity recommendation model, extracting high-order nonlinear feature interaction by using a fully-connected Deep neural network to obtain the output of the Deep part;
The outputs of the FM part and the Deep part are weighted and fused to obtain the final interest prediction scoreThe expression is:
;
Wherein,Representing a Sigmoid function;
Predictive scoring based on interestOrdering interest prediction scores of all commodities in descending order, and selecting the highest scoreIndividual article generation recommendation listAnd list the recommendationAnd displaying to a user.
The invention relates to a commodity recommending method based on big data analysis, which comprises the following steps of collecting behavior data and real-time data of store activities, dynamically adjusting a recommending list, and recording feedback of a user on recommended commodities:
Collecting behavior data and feedback data of a user in real time, and cleaning and preprocessing the behavior data and the feedback data;
according to the behavior data and feedback data of the user, dynamically calculating the recommendation priority score of each commodityThe expression is:
;
Wherein,Indicating that the user clicked on the merchandiseIs used for the total number of times of (a),Representing a user purchasing a commodityIs used for the total number of times of (a),Indicating that the user is on the commodityThe total dwell time on the page is such that,Representing user's pair of goodsIs used for the scoring of the (c),The base number representing the natural logarithm,AndRespectively representing the characteristic vector of store data and the characteristic vector of public opinion data,Representing user behavior time period characteristics.
In a second aspect, the invention provides a commodity recommendation system based on big data analysis, which comprises an acquisition module, a feature extraction module, a commodity recommendation module, an adjustment feedback module and a display module:
the acquisition module is in charge of acquiring relevant data of a user, preprocessing and integrating the data, and generating a recommended data set;
The feature extraction module is in charge of extracting key features from the acquired data, generating a user portrait and providing input for a recommendation model;
the commodity recommending module is used for constructing a recommending model based on the user portrait and commodity characteristics, predicting the interest score of the user on the commodity and generating a recommending list;
the adjustment feedback module is used for collecting user behavior data and store activity data in real time, dynamically adjusting a recommendation list, and recording and analyzing user feedback;
the display module is used for displaying the recommendation result to the user in a user-friendly mode and providing an interaction function.
In a third aspect, the invention provides a computer device comprising a memory and a processor, the memory storing a computer program, wherein the computer program when executed by the processor implements any of the steps of the big data analysis based commodity recommendation method according to the first aspect of the present invention.
In a fourth aspect, the present invention provides a computer readable storage medium having a computer program stored thereon, wherein the computer program when executed by a processor implements any step of the big data analysis based commodity recommendation method according to the first aspect of the present invention.
The commodity recommendation method based on big data analysis has the advantages that firstly, user behavior data, regional store data and commodity public opinion data are collected through a login module, a multi-dimensional recommendation data set is built, comprehensive data support is provided for recommendation, secondly, data accuracy and integrity are guaranteed through data cleaning, deduplication, anomaly correction and deficiency complement, then, user behavior characteristics, store activity characteristics and public opinion characteristics are extracted, comprehensive characteristic vectors are generated and used for building user portraits, depth learning model capture characteristic interaction is adopted based on the user portraits, commodity dynamic characteristics are combined to generate recommendation lists, recommendation results are further optimized through real-time collection of user feedback and dynamic adjustment of recommendation priority, the problems of poor data quality, response lag and insufficient individuation in the prior art are effectively solved, and the accuracy, the real-time performance and the user satisfaction of recommendation are remarkably improved.
Detailed Description
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present invention is not limited to the specific embodiments disclosed below.
Further, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic can be included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
Embodiment 1, referring to fig. 1 and 2, is a first embodiment of the present invention, and this embodiment provides a commodity recommendation method based on big data analysis, which includes the following steps:
the method comprises the steps of S1, collecting behavior data of a user on a platform and store data of an area where the user is located, collecting public opinion data of commodities from a social platform and a comment area, integrating the behavior data, the store data and the public opinion data to form a recommended data set, and preprocessing the recommended data set;
Combining the IP address of the user equipment with GPS positioning data to preliminarily acquire the geographic position of the user;
binding the user identification and the geographic position, and initially establishing a record;
According to the geographic position of the user, the store in the area where the user is located is selected from a store database, wherein the store geographic position, the on-shelf commodity and the activity information are included;
Acquiring behavior data of a user on a platform through a commodity recommendation platform, wherein the behavior data comprise click times, browsing times, searching times and purchasing times, and each behavior is attached with a time stamp and a behavior type;
behavior types including click, browse, search, and purchase;
According to the geographic position of the user, the store in the area where the user is located is selected from a store database, wherein the store geographic position, the on-shelf commodity and the activity information are included;
collecting public opinion data related to commodities from a social platform and a comment area, wherein the public opinion data comprises user evaluation, commodity heat and emotion tendency;
preprocessing refers to deleting repeated data, correcting abnormal data and complementing vacant data;
The behavior data, store data and public opinion data are integrated to form a recommended data set, and the expression is:
;
Wherein,Represent the firstA recommended data set for an individual user,Represent the firstThe behavior data of the individual user is stored,Representation and the firstStore data associated with the individual user,Representation and the firstGenerating a unique hash value for each piece of data in the recommended data set by using a hash table, comparing whether the hash value is repeated, deleting the repeated data after hash conflict detection, and thus realizing quick and efficient duplication removal;
Deleting repeated records in the behavior data, store data and public opinion data to ensure the uniqueness and accuracy of the data;
Detecting abnormal values in the recommended data set by using the quarter bit interval IQR, correcting the abnormal values, setting an abnormal threshold value, correcting data exceeding the abnormal threshold value by the average value of the similar data;
the threshold value for outlier correction is set to iqr×1.5, expressed as:
;
Wherein,Representing the third quartile of the data set,A first quartile representing a data set;
For example, in the user residence time data, ifThe time of the process is =60 seconds,Data exceeding 300+1.5× (300-60) =660 seconds is determined as an outlier and corrected by the same class user mean;
further, IQR is an index for measuring the degree of data dispersion in statistics, defined as the third quartile @, the fourth quartile @, the third quartile @) And the first quartile) Is the difference between (a):
;
iqr×1.5 is not directly taken as an anomaly threshold, but rather is 1.5 times the IQR slaveExtending upward to form a "tolerance range", beyond which data points are determined to be outliers, i.e. the outlier threshold isAnd 1.5 times IQR.
Predicting the value of missing data according to the values of K neighbor points most similar to the current data point by a K neighbor algorithm, and then carrying out missing completion on the data by a KNN interpolation method;
in the K-nearest neighbor algorithm with the missing value complement, the K value is set to 5-15, for example, through cross verification, when k=10, the interpolation precision (F1-score=0.92) is highest, and the calculation efficiency is optimal;
According to the invention, a multi-dimensional recommended data set is constructed through the steps of data integration and preprocessing, behavior feature extraction and public opinion data introduction, the accuracy and timeliness of recommendation are remarkably improved by combining a specific feature extraction method and algorithm optimization, the design of each step is based on the recommended core requirement, and meanwhile, the data processing efficiency and the recommendation effect are superior to those of the prior art through the technologies of Hash table de-weighting, IQR outlier correction, KNN interpolation and the like, and the technical scheme not only solves the difficulty of personalized recommendation of users, but also provides an innovative idea for dynamic adjustment of recommendation.
S2, extracting behavior characteristics from behavior data, extracting store characteristics from store data, and extracting public opinion characteristics from public opinion data;
Performing aggregation analysis on the behavior data, counting the behavior frequency of a user in a certain time window, extracting the distribution characteristics of behavior types, calculating the interest weight of the user on different types of commodities, extracting the preference characteristics of specific commodity types, extracting the time characteristics of the user behaviors, and extracting the behavior characteristics according to the formula:
;
Wherein,A feature vector representing the behavior data,The length of the time window is indicated,、、AndRespectively represent the time of the userThe number of clicks, searches, browses and purchases,The method is characterized in that the characteristic of the user behavior time period is represented, the value range is [0,1], and the behavior activity distribution of the user in one day is represented;
For example, the number of clicks, searches, browses, and purchases of the user C in the 7-day time window is [45,8,30,2], and the time activity h (t) =0.82 (night behavior ratio), then:
;
Clicking behavior average daily frequency =;
Search behavior average daily frequency =;
Browsing behavior average daily frequency =;
Buying behavior daily average frequency =;
Representing the activity distribution of the user in the day, and calculating by counting the frequency of the user's behavior in a specific time period (such as the morning, the afternoon and the night).
;
In a line data aggregation analysis, the time window lengthThe value of (2) is 7-30 days, for example, whenWhen t=30 days, the user's long-term interest profile (e.g., seasonal merchandise preference) may be covered;
The interest intensity of the user can be comprehensively reflected through independent statistics of clicking, searching, browsing and purchasing behaviors, for example, the user frequently searches for certain types of commodities but does not purchase the commodities, so that the type of commodities are potential demands;
Counting the number of shelves and classification distribution of commodities of the store in the area where the user is located, extracting matching features of commodity categories and inventory, extracting activity features (such as promotion discounts and time-limited discounts) of the store, calculating activity intensity, extracting distance features of geographic positions of the store and the user, measuring correlation of the store and the user through distance weights, and extracting a store feature extraction formula:
;
Wherein,A feature vector representing store data,Indicating the number of stores in the area where the user is located,Index variables representing the store's store locations,Represent the firstThe number of items on sale at a single store,Represent the firstThe preferential strength of the store activity is in the range of 0,1, the higher the value is, the greater the attraction of the store activity is,Representing the user and the firstThe straight line distance of the individual store,The distance weight function is used for calculating the weight of the influence of the distance on the recommendation, the value range is (0, 1), and the closer the distance is, the higher the weight is, and the commodity and the activity data of nearby stores are screened by combining the geographic position of the user, so that the recommendation of the commodity which exceeds the accessible range of the user is avoided, for example, the commodity of the store with the closer distance and the larger activity is more attractive, and the recommendation correlation is improved;
Counting the number of times of mention of the commodity in social media, extracting the heat characteristic of the commodity, analyzing comment content through an emotion analysis model, extracting the emotion tendency score of the commodity, and calculating the evaluation characteristic of the commodity by combining the heat of the commodity and the emotion tendency score, wherein a public opinion characteristic extraction formula is as follows:
;
Wherein,Feature vectors representing the public opinion data,Represents the total number of statistical public opinion data,Represent the firstThe forwarding amount of the public opinion data,Represent the firstEmotion tendency score of the strip public opinion data (calculated by emotion analysis model, range is [0,1 ]);
for example, 100 comments, and the total number of public opinion data in 100 comments=100, Average forwarding amount=350, Emotion score=0.85,
;
Total forwarding volume;
Let the forwarding amount of each public opinion be the same:
。
;
Total emotion score;
Let the emotion score of each public opinion be the same:
;
;
Due to=350 And=0.85:
;
;
;
The public opinion features combine the popularity of the commodity with the emotion evaluation, so that the commodity with high popularity and high evaluation can be preferentially recommended, for example, a commodity with high popularity and evaluated on social media is easier to attract users to click;
According to the invention, through extraction of behavior features, store features and public opinion features, a multi-dimensional feature vector model is constructed, and the design of each feature vector starts from the behavior, regional factors and market feedback of a user, so that the problems of insufficient individuation, regional deficiency and public opinion neglect in the traditional technology are solved, the time sensitivity features, distance weight functions and public opinion cross features are innovatively introduced, and the recommendation accuracy and user experience are remarkably improved.
S3, integrating the behavior features, the store features and the public opinion features into comprehensive feature vectors, and generating user portraits according to the comprehensive feature vectors;
The feature vector integration formula is:
;
Wherein,Representing the integrated comprehensive feature vector;
using PCA (principal component analysis) to performCompressing from high dimension (such as 200 dimension) to 128 dimension, retaining 95% of original information quantity, and improving calculation efficiency;
calculating a covariance matrix according to the comprehensive eigenvector, wherein the expression is as follows:
;
Wherein,The covariance matrix is represented by a matrix of covariance,Representing the total number of samples,The index of the sample is represented and,Represent the firstThe characteristic value of the individual samples is calculated,A mean vector representing the dataset;
Decomposing the characteristic value, wherein the expression is:
;
Wherein,A matrix of feature vectors is represented,Representing a diagonal matrix with dimensions ofDiagonal elements as eigenvaluesArranged in descending orderThe non-diagonal element is 0,Representing a transpose of the feature vector matrix;
before selectionThe feature vector (principal component) corresponding to the maximum feature value projects data into a low-dimensional space, main information is reserved, so that the cumulative variance contribution rate is more than or equal to 95%, and the expression is:
;
Wherein,Is the firstThe value of the characteristic is a value of,For the original dimension, for example, the original dimension is 200, when the first 128 main components are selected, the cumulative variance accounts for 95%, and the reasoning speed is improved by 40%;
Based on integrated feature vectorsConstructing user portraits of each user;
Comprehensive feature vectorEffectively combining user behavior, geographical region limitation and market feedback, for example, a user frequently browses a certain type of commodity (behavior characteristics), and a nearby store is promoting the commodity (store characteristics), and meanwhile, under the condition that the commodity is high in social media evaluation (public opinion characteristics), the commodity can be recommended preferentially, and the recommended hit rate and user satisfaction are improved remarkably;
The invention is characterized by the behaviorStore characteristicsAnd public opinion featuresIntegration into integrated feature vectorA multi-dimensional user portrait generation method is constructed, the problems of insufficient information dimension and strong data isolation in the traditional recommendation method are solved by the design of the comprehensive feature vector, the comprehensiveness and the dynamics of the user portrait are remarkably improved, in addition, the performance in regional recommendation and socialization recommendation is optimized by bringing geographic position and market feedback into the generation process of the user portrait, finally, the multi-dimensional high-precision user portrait can be generated in a shorter time, and a more accurate and real-time user demand expression means is provided for the recommendation method.
S4, constructing a commodity recommendation model based on the user portraits by using a machine learning algorithm, taking the user portraits as the input of the model, and outputting a commodity recommendation list by the commodity recommendation model in combination with public opinion heat of the commodity and store activity data in the area;
adopting a DeepFM model based on deep learning as a commodity recommendation model, and outputting interest prediction scores of users on commodities by combining user portraits, commodity characteristics and dynamic characteristics;
the commodity characteristics mainly refer to basic information (category, price and inventory) of commodities, and public opinion heat characteristics are calculated according to the following formula:
;
Wherein,Represents the public opinion popularity score of the commodity,Index variables representing the records of public opinion,Represent the firstThe forwarding amount of the public opinion data,Represent the firstThe emotion tendencies score of the strip public opinion data,A logarithmic function is represented and is used to represent,Represent the firstComment number of the strip public opinion data;
the public opinion heat threshold is set to be H not less than 0.65, for example, when H not less than 0.65, the click rate (CTR) of recommended commodities reaches 12.4%, and is improved by 143% compared with low heat commodities (CTR=5.1%);
And calculating the clicking behaviors of the recommended commodities by the CTR through statistics, wherein the display times are the total times of exposure in the recommendation list, and the clicking times are the times of actual clicking behaviors. For example, if a commodity is presented 1000 times and clicked 124 times, the CTR is 12.4% and the expression is:
;
the dynamic characteristics refer to real-time update conditions of store activity data and behavior data in the area, and the store activity characteristics of the area can be calculated by the following formula:
;
Wherein,Representing store activity intensity within the area;
the store distance weight function uses an exponential decay model:
;
for example, whenWhen the value of =5 km,;When the value of =10 km,The rule of 'closer distance, higher recommendation priority' is effectively reflected;
Depending on the combined effect of store activity strength and distance weight, when store activity is high and distance is close,The value increases;
Mapping commodity features and dynamic features to a low-dimensional dense vector space to generate feature embedded representations;
aiming at the feature intersection FM part of the commodity recommendation model, capturing the second-order interaction relation between input features by using a factorizer to obtain the output of the FM part;
Aiming at the nonlinear feature learning Deep part of the commodity recommendation model, extracting high-order nonlinear feature interaction by using a fully-connected Deep neural network to obtain the output of the Deep part;
The outputs of the FM part and the Deep part are weighted and fused to obtain the final interest prediction scoreThe expression is:
;
Wherein,Representing a Sigmoid function for normalizing the output to a [0,1] interval;
Predictive scoring based on interestOrdering interest prediction scores of all commodities in descending order, and selecting the highest scoreIndividual article generation recommendation listAnd list the recommendationThe recommendation list is displayed to the user, and the recommendation list generation formula is as follows:
;
Wherein,The list of recommendations is represented and,Indicating the number of recommended goods and,A threshold value (adjustable parameter, default value of 0.5) representing the interest score,Represent the firstThe number of articles to be manufactured is the same,Which represents the user's representation of the user,Representing a userFor commodityIs a predictive score of interest.
S5, collecting behavior data and real-time data of store activities, dynamically adjusting a recommendation list, and recording feedback of a user on recommended commodities;
collecting behavior data and feedback data of a user in real time, and cleaning and preprocessing the behavior data and the feedback data, wherein the preprocessing comprises de-duplication and outlier processing;
according to the behavior data and feedback data of the user, dynamically calculating the recommendation priority score of each commodityAdjusting a recommendation sequence based on a recommendation priority scoreThe expression is:
;
Wherein,Indicating that the user clicked on the merchandiseIs used for the total number of times of (a),Representing a user purchasing a commodityIs used for the total number of times of (a),Indicating that the user is on the commodityThe total dwell time on the page is such that,Representing user's pair of goodsThe higher the value, the higher the user's satisfaction with the product,The base number representing the natural logarithm,AndRespectively representing the characteristic vector of store data and the characteristic vector of public opinion data,Representing user behavior time period characteristics;
the weight distribution of the user behavior time period characteristics is as follows:
weights of 8:00-12:00 are 0.3 (e.g., user browsing behavior is more diffuse);
Weights of 12:00-18:00 are 0.5 (e.g., noon liveness boost);
Weights 0.8 (e.g., evening click conversion is 3.2 times the early morning period) from 18:00 to 22:00;
weights of 22:00-8:00 are 0.1 (e.g., user browsing behavior is more diffuse);
For example, using DeepFM model to integrate features of user CThe following are examples:
The FM part is used for calculating the second-order cross characteristic of the commodity F and the user C, and the weight is 0.32;
The expression of the second order cross feature is:
;
Wherein,Is a second order cross-over feature,AndFor the feature index to be used,As a dimension of the features,,As the hidden vector, the vector is a vector of the hidden vector,AndIn order to input the characteristic value of the characteristic,Is a hidden vector dot product.
Deep part extracts nonlinear relation through 3 layers of neural network (128- & gt 64- & gt 32) and outputs value 0.68.
Final scoring: (exceeding threshold 0.5, adding to the recommendation list);
dynamic adjustment stage, recommendation priority score calculation formula of commodity GWherein:
,,,
,,,
,,,
Then;
Recommendation priority score of commodity GAnd (285.44) indicating that the commodity G can be recommended preferentially in dynamic adjustment due to high public opinion enthusiasm, strong activity and positive feedback of users.
The invention dynamically calculates the recommendation priority score by collecting the user behavior data and the feedback data in real timeThe dynamic optimization of the recommendation method and the recommendation priority score are realized by combining the behavior characteristics, the store characteristics and the public opinion characteristicsThe formula design of the recommendation method comprehensively considers the user interest depth (click, purchase and stay time), satisfaction (score), regionality (store feature) and market feedback (public opinion feature), and improves the instantaneity and the accuracy of a recommendation result through time sensitivity adjustment.
The embodiment also provides a commodity recommendation system based on big data analysis, which comprises an acquisition module, a feature extraction module, a commodity recommendation module, an adjustment feedback module and a display module:
The acquisition module is in charge of acquiring relevant data of a user, preprocessing and integrating the data, and generating a recommended data set;
the feature extraction module is in charge of extracting key features from the acquired data, generating a user portrait and providing input for a recommendation model;
The commodity recommending module is used for constructing a recommending model based on the user portraits and commodity characteristics, predicting the interest scores of the users on commodities and generating a recommending list;
The adjustment feedback module is used for collecting user behavior data and store activity data in real time, dynamically adjusting a recommendation list, and recording and analyzing user feedback;
and the display module is used for displaying the recommendation result to the user in a user-friendly mode and providing an interaction function.
The embodiment also provides computer equipment, which is suitable for the condition of the commodity recommendation method based on big data analysis, and comprises a memory and a processor, wherein the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the commodity recommendation method based on big data analysis, which is provided by the embodiment.
The computer device may be a terminal comprising a processor, a memory, a communication interface, a display screen and input means connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.
The present embodiment also provides a storage medium having a computer program stored thereon, which when executed by a processor implements the commodity recommendation method based on big data analysis as proposed in the above embodiment, the storage medium may be implemented by any type of volatile or non-volatile storage device or combination thereof, such as a static random access Memory (Static Random Access Memory, SRAM for short), an electrically erasable Programmable Read-Only Memory (ELECTRICALLY ERASABLE PROGRAMMABLE READ-Only Memory, EEPROM for short), an erasable Programmable Read-Only Memory (Erasable Programmable Read Only Memory, EPROM for short), a Programmable Read-Only Memory (ROM for short), a magnetic Memory, a flash Memory, a magnetic disk, or an optical disk.
It should be noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention may be modified or substituted without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered in the scope of the claims of the present invention.