Detailed Description
Exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding, and are to be considered merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.
It should be noted that, the acquisition, storage, application, etc. of the personal information, etc. according to the embodiment of the present invention all conform to the rules of the related laws and regulations, and do not violate the popular regulations.
In the E-commerce searching scene, the behavior paths reflected by the searching behavior data of the user are very scattered and complex, and at present, the abnormal situation is determined mainly in the user searching process by describing the user transformation and the loss rate of key links, so that the whole searching behavior flow is optimized. User behavior analysis is performed only through conversion of a single link, so that the search shopping guide has low single rate and poor user experience.
In view of this, according to one aspect of an embodiment of the present invention, there is provided a user search behavior data analysis method.
Fig. 1 is a schematic diagram of a main flow of a user search behavior data analysis method according to an embodiment of the present invention. As shown in fig. 1, the user search behavior data analysis method according to an embodiment of the present invention includes the following steps S101 to S103.
Step S101, determining a user behavior path based on the acquired user search behavior data, wherein the user behavior path comprises a decision action and a decision state, the decision action comprises an operation behavior executed by a user, and the decision state comprises a page state in which the user is located.
The user search behavior data refers to related operation records generated when a user searches on an e-commerce platform or a search engine. In particular, the data includes search keywords entered by the user, items or pages clicked on, filtering operations performed, sorting actions, operations to join the shopping cart, and final purchasing actions, among others. A user behavior path refers to a complete behavioral link from an initial search operation to a final decision behavior (e.g., purchase or discard) by a user during a search and browsing process, and can be described by a series of decision actions and decision states. The decision action includes an operation action performed by the user, which may be a selection or operation made by the user on each node or in a state, such as clicking on a certain commodity or applying a certain screening condition. The decision state includes the page state in which the user is located, and may be the state in which the user is located after each operation, such as "browsing merchandise", "filtered merchandise" or "joining shopping cart", etc. The user behavior path can present the user's complete decision process throughout the shopping or searching process, showing the user's path from searching to final action.
And determining a user behavior path based on the acquired user search behavior data, arranging each operation of the user according to a time sequence, analyzing the relation between the operations, and identifying a behavior mode of the user to form a coherent behavior sequence. For example, from the search keyword of the user, it can be tracked whether the user narrows the search range through the filtering function, clicks on a certain commodity, checks other recommended commodities, and finally performs purchasing or discarding operations, and the sequence and association relationship of these steps form the behavior path of the user.
Step S102, markov modeling is carried out according to the decision action, a cost function is obtained, and the value degree of each decision state is determined based on the cost function.
The value degree of the decision state is the contribution degree of each user to the whole business target (such as platform transaction amount or user conversion rate) under different decision states. In the e-commerce platform or recommendation system, each decision state has its influence on the final conversion (purchasing behavior), and the importance of the decision state in the user behavior path is reflected by the value. A high value decision state means that the state is critical in user conversion or business index promotion.
In particular, the modeling process may be implemented by constructing a state transition matrix and calculating a cost function. Firstly, defining a state space (such as a search result page, a screening page, an item detail page and the like) and an action space (such as clicking, screening, sorting and the like) based on user behavior data, and then counting the probability of a user to transition to other decision states after executing certain decision actions under different decision states to form a state transition matrix. Meanwhile, the prize value is assigned to the state-action pair according to an instant prize function (e.g., increased conversion rate after clicking on the commodity or user residence time due to screening conditions). And iteratively calculating a cost function of each decision state by utilizing a Bellman equation, and reflecting the accumulated long-term benefits under the decision state, thereby determining the value degree of each decision state. The method can also dynamically optimize a cost function based on a reinforcement learning model, model a Markov decision process based on decision actions by using a reinforcement learning algorithm (such as deep Q learning), update the value estimation of decision state decision-action pairs by collecting user behavior paths in real time, gradually approach an optimal cost function according to current user behaviors and rewards, dynamically adapt to the change of the user behaviors, and more accurately determine the value degree of each decision state in the scene of insufficient user behavior data or rapid change of behavior patterns compared with static modeling by using a reinforcement learning mode.
By means of the method, behaviors of the user in different decision states can be systematically analyzed, valuable insight is provided for business decisions of the platform, for example, high-value states are optimized, or low-value states are adjusted, so that overall user conversion rate is improved.
And step S103, optimizing the searching shopping guide strategy according to the value degree of each decision state.
The searching shopping guide strategy comprises a strategy for providing personalized commodity recommendation and search result display for the user, and related commodities, services or commodity inventory adjustment are presented for the user by analyzing the search behavior and preference of the user so as to improve the shopping experience and conversion rate of the user. For example, the interest of the user in the merchandise is increased by optimizing shopping guide items in the user search results, which include category tiling, screening options, search drop-down, and the like. The category tiling is a way for intuitively displaying commodity classification, and the user can quickly browse and select interested commodity categories by tiling commodities in a grid or list form according to different categories. The screening options are tools for providing fine-grained searching for users, irrelevant commodities can be filtered through setting conditions, so that the users can more accurately position the commodities meeting requirements, specifically, the screening options comprise price intervals, brands, colors, specifications, scores and other dimensions, the users can dynamically adjust the conditions and update the searching results in real time, for example, on an electronic commerce platform, the users search for a 'notebook computer', and the screening of 'price below 5000 yuan', 'screen size: 14 inches' and the like can be selected. The search drop-down is a mode of recommending related words, commodities or phrases based on historical data or algorithms when a user inputs keywords, helps the user to complete input and find target contents more quickly, not only can the input efficiency be improved, but also the user can be guided to explore more commodities by providing diversified recommendation information to excite potential interests of the user, for example, the user inputs a mobile phone in a search box, and the drop-down can display a mobile phone shell, a mobile phone double eleven discount and the like. The reasonable layout of the shopping guide items can help the user to find the goods meeting the requirements more quickly, so that the shopping efficiency and satisfaction are improved.
In addition to adjusting the shopping guide, the search shopping guide strategy can also optimize the user experience and improve conversion through adjustment of merchandise inventory. Based on the value of each decision state, the inventory can be managed more accurately, and sufficient inventory is kept for the commodities related to the high-value decision state preferentially, so that the user demand can be met in time during the peak period of the user demand, and the loss caused by the shortage of the stock is avoided. Meanwhile, for commodities with low value or less searched by users, inventory investment can be properly reduced, inventory backlog risks are reduced, and supply chain management is optimized. Specifically, by analyzing the user's operational behavior under different decision states, such as clicking on the merchandise, viewing detail pages, joining shopping carts, etc., the platform can evaluate which merchandise is more focused by the user, and the cumulative value of these behaviors can be used to predict future merchandise needs and to prioritize the allocation of resources in inventory policies. For example, when the click rate and the shopping cart rate of a certain commodity are high and are related to a high-value decision state, the platform can predict that the commodity has higher sales conversion in the future and conduct inventory replenishment in advance, so that the commodity is ensured to be available in the decision process of a user. The search shopping guide strategy based on the decision state value degree is optimized, shopping guide items can be adjusted, user experience and the conversion rate of a platform can be improved through accurate inventory management and other modes, the search shopping guide strategy is adjusted more flexibly, the service experience of a user is improved, and meanwhile the success rate of commodity sales is effectively improved.
The embodiment of the invention determines the user behavior path based on the acquired user search behavior data, wherein the user behavior path comprises decision actions and decision states, performs relevance analysis on the decision actions, determines the value degree of each decision state based on a relevance analysis result, optimizes the search shopping guide strategy according to the value degree of each decision state, and can optimize the search shopping guide strategy by combining the user behavior path and performing value evaluation on the decision states among different decision actions, thereby improving the search shopping guide single rate and the user experience.
Optionally, markov modeling is performed according to the decision action to obtain a cost function, wherein the cost function comprises defining a model key element, the model key element comprises a state space, an action space, a state transition probability and a reward function, constructing a state transition matrix based on the model key element, and determining the cost function according to the decision action and the state transition matrix.
Key elements include state space, action space, state transition probabilities, rewards functions, and the like. The state space represents different decision states of the user on the platform, each corresponding to a specific context in which the user is located at a certain moment. The action space describes the action or behavior decisions that a user may take in different decision states, such as searching for new keywords, clicking on a certain item, ordering a list of items, or directly ordering. The state transition probabilities are used to describe the likelihood that a user will transition from one decision state to another. The reward function defines the benefits of the user's behavior, such as the amount of money the user successfully completed a purchase, placed, etc., and the platform may optimize the user's decision path based on the reward function to maximize the overall benefits of the platform.
After defining the key elements, a state transition matrix can be constructed based on the key elements, wherein the state transition matrix is a probability table describing the transition of the user behavior from one decision state to another decision state, and the next decision state possibly entered by the user under different decision actions can be predicted through the state transition matrix. The construction of the state transition matrix depends on behavior data of the user, the data comprise a series of operation steps from searching to clicking to purchasing by the user, and by analyzing the behavior tracks, a corresponding probability value can be allocated to each decision state transition, so that the matrix can accurately reflect dynamic changes of the behavior of the user. After the state transition matrix is constructed, a cost function may be determined based on the decision actions and the state transition matrix, and the cost function may be used to evaluate the long-range value of each decision state, reflecting the maximum jackpot that can be achieved by taking the best decision in that decision state. The cost function may be calculated by the bellman equation, which calculates the value of each decision state by recursion. The value of each decision state is further determined based on a cost function, and the high-value decision state represents a key node in a user behavior path and can influence user conversion and search efficiency. Based on these values, search shopping strategies may be optimized, such as enhancing user interaction at high value decision states, highlighting relevant screenings or recommendation tags, to improve user experience and conversion.
The embodiment of the invention can better understand the user behavior mode, optimize the searching and recommending strategies, and further improve the conversion rate and the user satisfaction.
Optionally, the method further comprises the steps of defining decision states in the state transition matrix as nodes, defining state transition probabilities exceeding a preset probability threshold in the state transition matrix as edges, constructing a transition path diagram based on the nodes and the edges, and extracting key loss paths from the transition path diagram so as to optimize the search shopping guide strategy based on the key loss paths.
In constructing the transition path graph, each decision state in the state transition matrix may be defined as a node, each node representing a particular stage in the operation of the user. Between these nodes, edges connecting the nodes are constructed according to the state transition probabilities in the state transition matrix, which reflect the probability of the user transitioning from one decision state to another. For optimal analysis, state transitions exceeding a preset probability threshold may be defined as edges, focusing on the most likely user behavior paths, thereby reducing data complexity. Based on the nodes and edges, a transition path diagram of the user behavior can be constructed, wherein the transition path diagram is a visual expression of the whole behavior path from the initial searching behavior to the final purchasing decision of the user, and the transition process and the corresponding probability of the user between different nodes are shown. Through analysis of the transfer path diagram, the behavior flow most frequently experienced by the user can be intuitively seen, and possible transfer paths of the user at different stages can be identified. And extracting a key loss path from the transfer path diagram, wherein the key loss path is a key decision node and a path which can be separated from a platform and terminate operation in the shopping process by a user, such as the user clicks a plurality of commodities and does not join a shopping cart, or the user joins the commodities into the shopping cart and does not complete the ordering. By identifying these critical churn paths, the platform can analyze the specific reasons that lead to the user abandoning the purchase. The search shopping guide strategy can be optimized aiming at the key loss paths, such as adjusting the display sequence of shopping guide items, the display mode of recommended content or optimizing the interactive experience of a user at a certain node, so that the loss probability is reduced. For example, if a user is found to be frequently lost in a particular item detail page, the content layout of the page may be optimized, promotional information added, or alternative items recommended. Through the optimization strategy based on the key loss path, the purchase conversion rate of the user can be effectively improved, and the overall benefit of the platform is improved.
Optionally, the method further comprises the steps of removing decision states in the state transition matrix one by one, updating the state transition matrix, and determining the descending amplitude of the user conversion probability based on the updated state transition matrix so as to optimize the search shopping guide strategy according to the descending amplitude of the user conversion probability.
In optimizing the search shopping guide strategy, a mode of removing decision states in the state transition matrix one by one can be adopted to evaluate the influence of each decision state on the conversion rate of the user. Firstly, aiming at the existing state transition matrix, the decision state in the state transition matrix is gradually and independently removed. After removing one decision state, the state transition matrix is updated accordingly. Based on the removed updated state transition matrix, the user transition probabilities in the overall user behavior path may be recalculated, which represents the overall likelihood that the user has completed a series of actions from search to purchase. The contribution degree of each decision state to the conversion rate can be evaluated by comparing the change of the user conversion probability before and after the removal, if the reduction degree of the user conversion probability is larger after a certain decision state is removed, the decision state is critical to the user conversion, otherwise, if the reduction degree is smaller, the influence of the state on the user behavior path is weaker. By analyzing these decreasing magnitudes, it can be determined which decision states play a key role in the final user conversion, and based on these analysis results, the search shopping guide strategy can be specifically optimized, for example, decision states that contribute more to the user conversion rate can be subjected to key display or pre-processing, so that the user can more easily contact shopping guide items in these states, and at the same time, for decision states that have less influence, it can be considered to reduce the recommended frequency or make adjustments thereof, so as to improve the overall efficiency of the shopping guide strategy. The shopping guide flow can be precisely optimized by gradually removing the state and observing the change of the user conversion rate, thereby improving the user conversion rate and the overall benefit.
Optionally, before determining the user behavior path based on the obtained user search behavior data, the method further comprises the steps of carrying out data cleaning on the obtained user search behavior data, constructing a user behavior operation log according to the cleaned user search behavior data, and screening a complete session set from the user behavior operation log based on the time similarity or category similarity so as to determine the user behavior path according to the user search behavior data corresponding to the complete session set.
The process of data cleansing includes removing invalid, duplicate or anomalous data, particularly data that is affected by robot operation, crawler interference or swipe, so that the retained data is more authentic and valid. According to the cleaned user search behavior data, a user behavior operation log is constructed, wherein the user behavior operation log is a data record organized according to individual users and comprises operation behaviors of browsing, clicking, purchasing and the like of each user, and each operation is attached with a timestamp and a behavior type mark, such as search keywords, screening clicks, commodity clicks, ordering and the like. Based on the time similarity or category similarity, a complete conversation set can be screened out, wherein the time similarity represents that continuous operations are closely separated in time and indicate that the operations belong to the same conversation, and the category similarity represents the commodity category correlation among the operations, and if a user intensively operates commodities of the same or similar category in a short time, the operations are classified into the same conversation. Through the two similarity standards, the continuous operation sequence of the user on the platform can be accurately identified, and a complete session set is formed. Based on the complete session set obtained by screening, the user behavior path is determined by utilizing the user search behavior data corresponding to the sessions, the complete session set represents all operation behaviors of the user in a certain requirement scene, the decision mode of the user in the scene is reflected, the user behavior path is determined by the data, and the accuracy of user behavior analysis can be improved.
Optionally, the searching shopping guide strategy comprises a strategy of optimizing a user search result by adjusting the display and the ordering of the shopping guide items, and the searching shopping guide strategy is optimized according to the value degree of each decision state, wherein the searching shopping guide strategy comprises the steps of leading the shopping guide items corresponding to the decision states with the value degree exceeding a preset threshold value, and removing the shopping guide items corresponding to the decision states with the value degree not exceeding the preset threshold value.
And (3) pre-processing shopping guide items corresponding to the high-value decision state, namely, preferentially displaying contents related to the high-value decision state when a user searches, for example, if the screening function has a large contribution to the conversion rate of the user, the screening options can be designed to be more obvious and convenient, and even a part of screening conditions are preselected according to the habit of the user by default. Accordingly, for decision states with lower value, the corresponding shopping guide items can be selected to be simplified or removed, so that the interference and unnecessary selection of the user are reduced, and the user is helped to find the required commodity more quickly. In addition, the shopping guide strategy can be dynamically adjusted by using machine learning methods such as reinforcement learning, the value of the decision state is updated in real time by continuously learning the real-time operation behavior data of the user, and the ordering and presenting modes of shopping guide items are automatically adjusted according to the change, so that the user experience and the nutritive conversion rate of the platform are effectively improved.
FIG. 2 is a flow chart of a method for analyzing user search behavior data according to a preferred embodiment of the present invention, wherein as shown in FIG. 2, a search shopping guide strategy is optimized based on user behavior data, a preparation data set is obtained by first recording user session granularity, and specific search behaviors and interactive operations of a user on an e-commerce platform are collected to form a complete behavior data set. The method comprises the steps of constructing a model by using a Markov decision process, and firstly defining core elements in the model, wherein the core elements comprise decision states, decision actions, discount factors (for balancing the value of current and future decisions) and rewarding functions (for evaluating the rewards of users in various operations), and the elements together form a mathematical model capable of simulating the decision process of the users. After the model is built, a state transfer function is generated, the state transfer function describes the probability of a user transferring from one decision state to another decision state, and in the initial model generation stage, the state transfer function is still in a preliminary form and needs to be further optimized and perfected later. In the strategy evaluation stage, the state transfer function generated before is utilized to calculate the cost function of each decision state and decision action, so that the effectiveness of different strategies is evaluated, namely, the strategy which can bring the maximum benefit is searched, and the shopping guide tool is ensured to maximally improve the conversion rate of the user. A critical churn path is identified based on the state transition matrix. In the process that the user shifts from one decision state to another decision state, if the user losses on some paths are more, the key loss paths need to be paid attention to, key nodes can be found out by analyzing the key loss paths, and targeted optimization measures such as adding supplementary supplies of some commodities or optimizing related search words can be adopted to reduce the losses of the user. And meanwhile, the state transition matrix can be utilized to inquire the state transition which can bring high benefit, N inquires with the highest benefit are selected from the state transition matrix, the inquires are regarded as the next operation which is most helpful to the user, and the user can be effectively guided to complete purchase conversion by recommending the inquires for the user. The first N queries are used as shopping guide tool recommendation guide items of the current node, so that a user is helped to find interesting commodities more quickly, and optimization and promotion of a platform shopping guide strategy are finally achieved.
FIG. 3 is a schematic flow chart of a data preparation stage according to a preferred embodiment of the present invention, as shown in FIG. 3, a data set is obtained from an e-commerce platform, the data set is mainly a browse click log and a purchase record of a user, meaningless browse click data such as cheating actions of a robot and a crawler are removed, order form removal and refreshing data are removed, a user action operation log o= < uuid, query, event_id, time > is obtained according to record combination, wherein uuid represents the user, query represents a search keyword, event_id represents action events (such as screening click, sorting click, commodity click, purchasing order, etc.) of various clicks and purchases of the user, time represents user action event trigger time, user session is then identified, log data is organized, session S= { o1,o2, ...,oi }, all execution users of operation oi are identical, then a new session is started for the same service represented by the same time sequence of operation log is judged according to the time, if the high correlation category of query is consistent, that oi +1 belongs to session S, and the new session is started for the same service record of the same operation sequence according to the same meaning of the user.
Specifically, search logs for users are collected and consolidated from an e-commerce platform, which logs contain browsing, clicking and purchasing records for users. During the collection process, the log data is filtered to exclude meaningless operations such as robot and crawler behaviors and abnormal order data such as a billing behavior. The operation logs are grouped according to a User Unique Identifier (UUID), ensuring that each group of log data belongs to the same user. After grouping, the operation logs of each user are ordered chronologically so that the time series of user behavior can be better analyzed. After the sorting is completed, the operation behaviors of the users are analyzed one by one. First, it is determined whether the current operation is the first action of the user, and if so, a new session is created. If not the first operation, it is checked whether the operation is highly relevant to the search keyword of the previous operation, if so, both are considered to belong to the same session, otherwise a new session is created. Within the same session, it is further determined whether the event type operated by the user is a purchase event. If it is a purchase event, the session is deemed to be ended and the complete session is added to the final dataset. Otherwise, continuing to record the behavior in the session until the session end condition is reached or a new purchase behavior occurs. In this way, the system can effectively organize the operation records of the user on the same type of requirements into a session, and clear structured data is provided for subsequent analysis.
In the preferred embodiment of the invention, in the process of constructing a model by utilizing a Markov decision process, a search session process between a user and a search engine is that the user inputs ' men ' shoes as queries in a search box and clicks a ' search ' button, the search engine executes a search request once and displays commodity pages of keywords ' men ' shoes ', the user browses commodity display pages, can click on some commodity entering detail pages, the user finds out unsatisfactory commodities, can filter and rearrange the pages through a product module such as screening and sorting, can refine requirements through a product module such as category tiling, middle interleaving and the like, or can actively rewrite and re-input new search keywords, the search engine receives the request, executes a second search request and displays commodity pages corresponding to the second search, and the search session is finished when the user purchases a certain commodity or browses for a plurality of times.
Firstly, modeling context information and user behaviors in the process of searching session, defining concepts such as search result pages, user behavior events, success conversion rate and the like, which are the basis for defining states and state transition relations. Defining query access history, let q be the query of a search session, for an initial time step t=0, the corresponding initial query access path is h0=q, for any other time step t > =1, the corresponding access path is ht= (ht-1, qt), where ht-1 is the access path of step t-1, qt is the access query of step t, different users choose to search, purchase or leave at different time steps, and purchase and leave are two termination states, so that conversion probability and abandon probability can be defined based on the termination states. Defining the probability of the turn-into-turn, for any path node in a search session, ht (t > 0), letting B (ht) represent the random event of the purchase behavior of the user after observing ht, namely, the probability of the turn-into-turn of ht, denoted as B (ht), namely, the probability of the occurrence of event B (ht) under ht. Defining the abandon probability, for any path node ht (t > 0) in a search session, let L (ht) represent a random event that a user leaves the search session after observing ht, and then marking the abandon probability of ht as L (ht), namely the probability that the event L (ht) occurs under ht. Defining a re-search probability, for any path node ht (t > 0) in a search session, letting C (ht) represent a random event of a user initiating a secondary search after observing ht, and marking the re-search probability of ht as C (ht) is the probability of occurrence of event C (ht) under ht. Defining a decision process of a search session, wherein the decision process of the search session can be formally defined as M= < T, H, S, A, R, P >, and T in the tuple is the maximum decision step number of the search; for searching all covered query access histories in session, wherein Ht is the set of all possible queries at t time);In order to be a state space,Is a non-terminating set of states that contains all continuing sessions,AndA set of termination states containing all the turn-on and turn-off events, respectively; For action interval, all product modules triggering re-search (generalized) are included, including but not limited to a pull-down module (pull-down words, history searches, dark marks, search findings, hot search charts), active searches, sorting classes (sales sorting, price sorting, comment sorting), screening classes (Beijing dong logistics, clapping second hands, market same money, new products, etc.), re-search classes (class tiling, middle labels, related searches, merchant card tags, etc.), users trigger one re-search after clicking the product module, part of the product modules do not modify original search words, in order to distinguish different search pages, such search times manually correct search words into original query+product module names (e.g. men's shoes+price sorting) in data preprocessing, R is a reward function, which defines a reward function to promote transactions between users and sellers as much as possible for any time step #) In observing that the user purchases goods with an average probability of b (ht+1) on a certain search page ht+1, although different users select different goods to purchase, from the statistical point of view, the price of the goods corresponding to the conversion time occurring on ht+1 will necessarily follow a specific distribution, m (ht+1) represents the expectations of the goods on the search page ht+1, and the average relevance coefficient of the goods on the search result page and the search word is represented by m (ht+1), if the relevance coefficient is smaller than 0.5, the relevant goods supply is considered to be lacking under the search word, the user experience is affected to some extent, the average price of the goods exposed to be given with a negative number at this time, namely the current search page is defined, the user executes action a on state C (ht) to be transferred to any stateThe prize definition of (c) is as follows:
Wherein,Is thatThe corresponding end state of the transaction conversion event occurring above can be seen from the definition of the reward function, the positive reward can be obtained only when the transaction occurs, the negative reward can be obtained when the supply is insufficient, in other cases, the obtained reward value is zero, the expected price of the transaction of any search query result page is generally unknown, and in specific application, the actual transaction price of each time can be used as the reward signal of the corresponding transaction event.
Specifically, the algorithm design process makes M= < H, S, A, R, P > a search session decision process,Representing a set of policies, elements thereofThe calculation mode of the accumulated prize value obtained under any state s is constructed and can be expressed by the following state value function formula:
Simultaneous definition of action cost functionsIndicating compliance with policies at MDPThe expected return from performing action a on the current state s:
Wherein V(s) is a function of the value of state s,Calculating a value function to be converged according to a historical strategy to obtain a converted state transition matrix as a discount factor, and calculating action cost functions for the states sAnd outputting all the subsequent state expected returns of the state s, generating an optimal recommended item according to the expected return, and taking the optimal recommended item as an alternative item (category tiling, middle label) of a shopping guide tool as a page prefetched or recommended to a user, so that the purchase path of the user is further shortened, and the platform flow conversion rate is improved. And extracting the behavior states and paths required to be experienced from each row of states to the loss state based on the obtained transition path diagram to obtain the key loss path of each behavior state, wherein the key loss path causing the user loss can be effectively identified, and simultaneously the state value of each state and the removal effect can be combined, the removal effect refers to the change value of the sum of the probabilities on all paths from the initial state to the conversion state after removing a certain state, and the removal effect actually reflects the degree of reduction of the overall conversion rate of the system after removing the channel. Furthermore, the service operation can pertinently optimize goods supply, shopping guide tools and search results in cooperation with a search algorithm, so that the conversion efficiency of flow and the user experience are improved.
According to a second aspect of the embodiment of the invention, a shopping guide strategy optimization device is provided.
Fig. 4 is a schematic diagram of main modules of the shopping guide strategy optimization device according to an embodiment of the present invention. As shown in fig. 4, a shopping guide strategy optimization device 400 includes:
The path determining module 401 is configured to determine a user behavior path based on the obtained user search behavior data, where the user behavior path includes a decision state and a decision action;
a value determining module 402, configured to perform markov modeling on the decision actions to obtain a cost function, and determine a value degree of each decision state based on the cost function;
the policy optimization module 403 is configured to optimize the search shopping guide policy according to the value degree of each decision state.
Optionally, the value determination module 402 is further configured to:
Defining model key elements, wherein the model key elements comprise a state space, an action space, a state transition probability and a reward function;
constructing a state transition matrix based on the key elements of the model;
a cost function is determined based on the decision action and the state transition matrix.
Optionally, the apparatus 400 further includes a churn path module, where the churn path module is configured to:
defining decision states in the state transition matrix as nodes, and defining state transition probabilities exceeding a preset probability threshold in the state transition matrix as edges;
constructing a transfer path diagram based on the nodes and the edges;
And extracting a key loss path from the transfer path diagram so as to optimize the searching shopping guide strategy based on the key loss path.
Optionally, the apparatus 400 further comprises a removal effect module for:
and removing decision states in the state transition matrix one by one, updating the state transition matrix, and determining the descending amplitude of the user conversion probability based on the updated state transition matrix so as to optimize the search shopping guide strategy according to the descending amplitude of the user conversion probability.
Optionally, the apparatus 400 further includes a data cleansing module, where the data cleansing module is configured to:
Data cleaning is carried out on the obtained user search behavior data, and a user behavior operation log is constructed according to the cleaned user search behavior data;
And screening the complete session set from the user behavior operation log based on the time similarity or category similarity, so as to determine a user behavior path according to the user search behavior data corresponding to the complete session set.
Optionally, the search shopping guide strategy comprises a strategy for optimizing user search results by adjusting presentation and ordering of shopping guide items, and the strategy optimization module 403 is further configured to:
Leading the shopping guide items corresponding to the decision states with the value degree exceeding the preset threshold value, and removing the shopping guide items corresponding to the decision states with the value degree not exceeding the preset threshold value.
The specific implementation of the shopping guide policy optimizing apparatus of the present invention has been described in detail in the above method of optimizing the shopping guide policy, and therefore, the description thereof will not be repeated here.
According to a third aspect of the embodiment of the present invention, there is provided an electronic device including one or more processors, and a storage device for storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the method provided by the first aspect of the embodiment of the present invention.
According to a fourth aspect of embodiments of the present invention, there is provided a computer readable medium having stored thereon a computer program which, when executed by a processor, implements the method provided by the first aspect of embodiments of the present invention.
According to a fifth aspect of embodiments of the present invention, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the method provided by the first aspect of embodiments of the present invention.
Fig. 5 illustrates an exemplary system architecture 500 in which the shopping guide policy optimization method or device of embodiments of the present application may be applied.
As shown in fig. 5, the system architecture 500 may include terminal devices 501, 502, 503, a network 504, and a server 505. The network 504 is used as a medium to provide communication links between the terminal devices 501, 502, 503 and the server 505. The network 504 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.
A user may interact with the server 505 via the network 504 using the terminal devices 501, 502, 503 to receive or send messages or the like. Various communication client applications may be installed on the terminal devices 501, 502, 503, such as shopping class applications, web browser applications, search class applications, instant messaging tools, mailbox clients, social platform software, etc. (by way of example only).
The terminal devices 501, 502, 503 may be a variety of electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablets, laptop and desktop computers, and the like.
The server 505 may be a server providing various services, such as a background management server (by way of example only) providing support for shopping-type websites browsed by users using the terminal devices 501, 502, 503. The background management server may analyze and process the received data such as the shopping guide policy optimization request, and feed back the processing result (e.g., the shopping guide policy optimization result—only an example) to the terminal device.
It should be noted that, the shopping guide policy optimization method provided in the embodiment of the present application is generally executed by the server 505, and accordingly, the shopping guide policy optimization device is generally disposed in the server 505.
It should be understood that the number of terminal devices, networks and servers in fig. 5 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.
Referring now to FIG. 6, there is illustrated a schematic diagram of a computer system 600 suitable for use in implementing an embodiment of the present application. The terminal device shown in fig. 6 is only an example, and should not impose any limitation on the functions and the scope of use of the embodiment of the present application.
As shown in fig. 6, the computer system 600 includes a Central Processing Unit (CPU) 601, which can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 602 or a program loaded from a storage section 608 into a Random Access Memory (RAM) 603. In the RAM603, various programs and data required for the operation of the computer system 600 are also stored. The CPU601, ROM602, and RAM603 are connected to each other through a bus 604. An input/output (I/O) interface 605 is also connected to bus 604.
Connected to the I/O interface 605 are an input section 606 including a keyboard, a mouse, and the like, an output section 607 including a Cathode Ray Tube (CRT), a liquid crystal credit authorization query processor (LCD), and the like, and a speaker, and the like, a storage section 608 including a hard disk, and the like, and a communication section 609 including a network interface card such as a LAN card, a modem, and the like. The communication section 609 performs communication processing via a network such as the internet. The drive 610 is also connected to the I/O interface 605 as needed. Removable media 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is installed as needed on drive 610 so that a computer program read therefrom is installed as needed into storage section 608.
In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication portion 609, and/or installed from the removable medium 611. The above-described functions defined in the system of the present application are performed when the computer program is executed by a Central Processing Unit (CPU) 601.
The computer readable medium shown in the present application may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium may include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of a computer-readable storage medium may include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present application, however, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, with the computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The units involved in the embodiments of the present application may be implemented in software or in hardware. The described units may also be provided in a processor, which may be described as, for example, a processor comprising a path determination module, a value determination module and a policy optimization module. The names of these modules do not constitute limitations on the module itself in some cases, and for example, the path determination module may be described as "a module for determining a user behavior path based on acquired user search behavior data".
As a further aspect, the application also provides a computer readable medium which may be comprised in the device described in the above embodiments or may be present alone without being fitted into the device. The computer readable medium carries one or more programs which, when executed by the device, cause the device to include determining a user behavior path based on the obtained user search behavior data, wherein the user behavior path includes a decision action and a decision state, the decision action includes an operation action performed by the user, the decision state includes a page state in which the user is located, performing Markov modeling according to the decision action to obtain a cost function, determining a value degree of each decision state based on the cost function, and optimizing a search shopping guide strategy according to the value degree of each decision state.
The computer program product provided by the embodiment of the invention comprises a computer program, and the computer program realizes the shopping guide strategy optimization method in the embodiment of the invention when being executed by a processor.
The technical scheme provided by the embodiment of the invention has the advantages or beneficial effects that the user behavior path is determined based on the acquired user search behavior data, wherein the user behavior path comprises decision actions and decision states, the decision actions comprise operation actions executed by a user, the decision states comprise page states where the user is located, markov modeling is carried out according to the decision actions to obtain a cost function, the value of each decision state is determined based on the cost function, the search shopping guide strategy is optimized according to the value of each decision state, and therefore, the search shopping guide strategy can be optimized by carrying out value evaluation on the decision states among different decision actions in combination with the user behavior path, so that the single rate of search shopping guide and the user experience are improved.
The above embodiments do not limit the scope of the present application. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives can occur depending upon design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of the present application.
It should be noted that, in the technical solution of the present disclosure, the acquisition, storage, application, etc. of the related personal information of the user all conform to the rules of the related laws and regulations, and do not violate the popular regulations of the public order.