User behavior processing method and apparatus, electronic device, and computer readable medium

Abstract

The application discloses a user behavior processing method and device, electronic equipment and computer readable medium, relates to the technical field of big data, and particularly relates to the technical field of data analysis and mining. The method comprises the following steps: receiving a user behavior processing request, obtaining a corresponding website identifier and a user identifier, and obtaining a corresponding website log based on the website identifier; determining a user click path corresponding to the user identifier based on the website log, obtaining the conversion rate of each node in the user click path; determining a target node in each node according to the conversion rate, and then calling a frequent pattern mining program to determine a frequent path according to the target node; inputting the frequent path into a prediction model to generate a predicted user click path, and then adjusting the website structure corresponding to the website identifier based on the predicted user click path. Thus, the associated frequent path is maximally mined, the generality and accuracy of the method for processing the user behavior are improved, and the overall processing efficiency of the user behavior analysis is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data, in particular to the technical field of data analysis and mining, and especially to a user behavior processing method and device, an electronic device and a computer readable medium. BACKGROUND

[0002] The basic data of log-based user behavior analysis mainly comes from Web logs recording user browsing behavior. Through analysis and feature extraction of Web log texts, the analysis and prediction of user behavior are realized. However, the existing method for analyzing and predicting user behavior by analyzing and extracting features from Web log texts is not universal and has low overall analysis efficiency.

[0003] In the process of implementing the present application, the inventors have found that the prior art at least has the following problems:

[0004] The existing method for analyzing and predicting user behavior by analyzing and extracting features from Web log texts has a long analysis period, low accuracy, is not universal, and has low overall processing efficiency for user behavior analysis. SUMMARY

[0005] Therefore, the embodiments of the present application provide a user behavior processing method and device, an electronic device and a computer readable medium, which can solve the problem that the existing method for analyzing and predicting user behavior by analyzing and extracting features from Web log texts is not universal and has low overall analysis efficiency.

[0006] To achieve the above object, according to an aspect of the embodiments of the present application, a user behavior processing method is provided, comprising:

[0007] receiving a user behavior processing request, obtaining a corresponding website identifier and a user identifier, and obtaining a corresponding website log based on the website identifier;

[0008] determining a user click path corresponding to the user identifier based on the website log, and obtaining a conversion rate of each node in the user click path;

[0009] determining a target node in each node according to the conversion rate, and then calling a frequent pattern mining program to determine a frequent path according to the target node;

[0010] inputting the frequent path into a prediction model to generate a predicted user click path, and then adjusting a website structure corresponding to the website identifier based on the predicted user click path.

[0011] Optionally, determining a target node in each node according to the conversion rate comprises:

[0012] determining a node in each node whose conversion rate is less than a preset threshold;

[0013] determine a node with a conversion rate less than a preset threshold as a target node in the nodes.

[0014] Optionally, the frequent pattern mining procedure is invoked to determine a frequent path according to the target node, including:

[0015] The target node session corresponding to the target node is screened out.

[0016] The conversion rate of the target node is determined, and the conversion rate of the target node is determined as a support threshold.

[0017] The target node session and the support threshold are input into the frequent pattern mining procedure to obtain a frequent path.

[0018] Optionally, the method further includes:

[0019] In response to the frequent path being empty, the support threshold is adjusted based on a preset threshold gradient until the obtained frequent path is non-empty.

[0020] Optionally, the frequent path is obtained, including:

[0021] Each candidate path is generated according to the target node session.

[0022] The support of each candidate path is determined.

[0023] The candidate path corresponding to the support greater than the support threshold is determined as the frequent path.

[0024] Optionally, the predicted user click path is generated, including:

[0025] The frequent path is converted into a frequent path vector.

[0026] The user click distribution is predicted based on the frequent path vector, and then the predicted user click path is generated.

[0027] Optionally, the user click path corresponding to the user identifier is determined based on the website log, including:

[0028] The session identifier corresponding to the user identifier in the website log is obtained.

[0029] The corresponding session data is obtained based on the session identifier.

[0030] The user click path corresponding to the user identifier is constructed based on the session data.

[0031] Optionally, the website structure corresponding to the website identifier is adjusted based on the predicted user click path, including:

[0032] The current access path corresponding to the user identifier is obtained, and the access probability of each node in the predicted user click path is determined.

[0033] determine the node corresponding to the maximum access probability, and then generate a predicted website structure according to the current access path and the node corresponding to the maximum access probability;

[0034] obtain an expected website structure, and then update the expected website structure based on the predicted website structure.

[0035] In addition, the present application also provides a user behavior processing device, comprising:

[0036] a receiving unit configured to receive a user behavior processing request, obtain a corresponding website identifier and a user identifier, and obtain a corresponding website log based on the website identifier;

[0037] an obtaining unit configured to determine a user click path corresponding to the user identifier based on the website log, and obtain a conversion rate of each node in the user click path;

[0038] a frequent path determining unit configured to determine a target node in each node according to the conversion rate, and then call a frequent pattern mining program to determine a frequent path according to the target node;

[0039] an adjusting unit configured to input the frequent path into a prediction model to generate a predicted user click path, and then adjust a website structure corresponding to the website identifier based on the predicted user click path.

[0040] Optionally, the frequent path determining unit is further configured to:

[0041] determine a node with a conversion rate less than a preset threshold value in each node;

[0042] determine the node with the conversion rate less than the preset threshold value as the target node in each node.

[0043] Optionally, the frequent path determining unit is further configured to:

[0044] obtain a target node session corresponding to the target node through screening;

[0045] determine the conversion rate of the target node, and determine the conversion rate of the target node as a support threshold value;

[0046] input the target node session and the support threshold value into the frequent pattern mining program to obtain the frequent path.

[0047] Optionally, the adjusting unit is further configured to:

[0048] in response to the frequent path being empty, adjust the support threshold value based on a preset threshold gradient until the obtained frequent path is non-empty.

[0049] Optionally, the frequent path determining unit is further configured to:

[0050] generate each candidate path according to a target node session;

[0051] determine a support degree of each candidate path;

[0052] determine the candidate path corresponding to the support degree greater than the support degree threshold as a frequent path.

[0053] Optionally, the adjusting unit is further configured to:

[0054] convert the frequent path into a frequent path vector;

[0055] predict a user click distribution based on the frequent path vector, and further generate a predicted user click path.

[0056] Optionally, the obtaining unit is further configured to:

[0057] obtain a session identifier corresponding to the user identifier in a website log;

[0058] obtain corresponding session data based on the session identifier;

[0059] construct a user click path corresponding to the user identifier based on the session data.

[0060] Optionally, the adjusting unit is further configured to:

[0061] obtain a current access path corresponding to the user identifier, and determine an access probability of each node in the predicted user click path;

[0062] determine a node corresponding to a maximum access probability, and further generate a predicted website structure according to the current access path and the node corresponding to the maximum access probability;

[0063] obtain an expected website structure, and further update the expected website structure based on the predicted website structure.

[0064] In addition, the present application further provides a user behavior processing electronic device, comprising: one or more processors; a storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the user behavior processing method as described above.

[0065] In addition, the present application further provides a computer readable medium having a computer program stored thereon, the program is executed by a processor to implement the user behavior processing method as described above.

[0066] To achieve the above object, according to another aspect of the embodiments of the present application, a computer program product is provided.

[0067] The computer program product of the embodiments of the present application comprises a computer program, and the program is executed by a processor to implement the user behavior processing method provided by the embodiments of the present application.

[0068] One embodiment of the above application has the following advantages or benefits: the present application receives a user behavior processing request, obtains a corresponding website identifier and a user identifier, obtains a corresponding website log based on the website identifier, determines a user click path corresponding to the user identifier based on the website log, obtains a conversion rate of each node in the user click path, determines a target node in each node according to the conversion rate, and then calls a frequent pattern mining program to determine a frequent path according to the target node; the frequent path is input into a prediction model to generate a predicted user click path, and then the website structure corresponding to the website identifier is adjusted based on the predicted user click path. The node with a low single-step conversion rate found by the funnel model is taken as the target node, and the number of clicks of the target node is taken as the threshold value input into the frequent pattern mining program, so as to maximize the mining of the associated frequent path. By using the method combining the funnel model and the frequent pattern mining, the conversion rate of the website is analyzed, the big data processing technology is used to quickly mine the massive click path data, the analysis period is shortened, and the prediction model based on the neural network is used to predict the user click path, so as to provide a reference basis for website operators to optimize the information structure of the key page, improve the generality and accuracy of the method for processing the user behavior, and improve the overall processing efficiency of the user behavior analysis.

[0069] The further effects of the above-mentioned non-conventional optional mode will be described in the following combined with the specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0070] The accompanying drawings are used to better understand the present application, and do not constitute an improper limitation on the present application. Among them:

[0071] Figure 1 is a schematic diagram of the main process of the user behavior processing method according to the first embodiment of the present application;

[0072] Figure 2 is a schematic diagram of the main process of the user behavior processing method according to the second embodiment of the present application;

[0073] Figure 3 is a schematic diagram of the application scenario of the user behavior processing method according to the third embodiment of the present application;

[0074] Figure 4 is a schematic diagram of the main unit of the user behavior processing device according to the embodiment of the present application;

[0075] Figure 5 is an exemplary system architecture diagram to which the embodiment of the present application can be applied;

[0076] Figure 6 is a structural schematic diagram of a computer system suitable for implementing the terminal device or server of the embodiment of the present application. DETAILED DESCRIPTION

[0077] Exemplary embodiments of the present application are described herein with reference to the accompanying drawings, which are meant to be exemplary in nature, and in which various specific details are set forth in order to provide a thorough understanding of the application. However, it should be understood that various embodiments of the present application can be practiced without many of the details given below. Furthermore, well-known functions or constructions are not shown or described in detail since it would be

[0078] Figure 1 is a schematic diagram of the main flow of the user behavior processing method according to the first embodiment of the present application, as shown in Figure 1 the user behavior processing method comprises:

[0079] In step S101, a user behavior processing request is received, a corresponding website identifier and a user identifier are obtained, and a corresponding website log is obtained based on the website identifier.

[0080] In the present embodiment, the execution subject of the user behavior processing method (for example, it can be a server corresponding to a user behavior analysis system) can receive a user behavior processing request through wired connection or wireless connection. The user behavior processing request can be, for example, a request for analyzing and predicting the user's browsing behavior. The website identifier and the user identifier can be carried in the user behavior processing request. The website identifier is used to represent which website the user behavior analysis is performed for. The user identifier is used to represent which user the user behavior analysis is performed for. After obtaining the website identifier, the execution subject can obtain the corresponding website log based on the website identifier, and more specifically, the execution subject can also filter out the website log corresponding to the user identifier from the obtained website log for subsequent specific analysis.

[0081] In step S102, the user click path corresponding to the user identifier is determined based on the website log, and the conversion rate of each node in the user click path is obtained.

[0082] In the present embodiment of the present application, each node in the user click path corresponds to a webpage, that is, each node corresponds to a webpage.

[0083] The user's jump structure between webpages forms a graph structure, and the webpage is the vertex and the browsing jump is the edge. The execution subject can filter out the click webpage containing the user identifier from the website log, and each webpage containing the user identifier filtered out is taken as a node and connected in turn, thereby forming the user click path.

[0084] The construction of the user click path includes the following steps: web site log cleaning, user identification, session identification, and user click path restoration. Through processing the web site log and solving the scalability problem of mass data processing, the system can realize horizontal expansion at any time to cope with the rapid growth of data. In the process of web log preprocessing, the execution subject can use the MapReduce distributed computing framework to process a large amount of unstructured web log data. The search, click and other browsing behaviors of the users of the e-commerce website are recorded in the web log (i.e., website log) of the website. The web log of the website contains logs of all resources of the website and access of web crawlers, which are usually unstructured or semi-structured data. User identification: In order to more accurately identify users, the execution subject can use the cookie technology provided by the browser to realize user identification. Session identification: The web server generates a session for the browser to identify the user's interaction with the website for a period of time. Another session will be generated if the operations between two times of the browser exceed a certain threshold (usually 30 minutes). The website uses this session technology to save the login state of the user. In order to improve the access experience of the user who is not logged in, the e-commerce website does not use the session technology provided by the web server to distinguish the user's session, but this mechanism can provide a reference for session identification. Through web log preprocessing (log data cleaning and summarizing), user identification and session identification, specific data is formed, all click paths of the browser are constructed through session data, and the click times of the nodes, the click order of the nodes, the position and number of parent nodes (backtracking times), and other data are recorded. The user identification corresponding to the user click path is generated according to the click times of the nodes, the click order of the nodes, the position and number of parent nodes, and other data, which serves as the basis for the next step of analyzing user behavior. The conversion rate of each node in the user click path is obtained (conversion rate = (conversion times / click volume) x 100%).

[0085] In step S103, the target node in each node is determined according to the conversion rate, and a frequent pattern mining program is called to determine the frequent path according to the target node.

[0086] Specifically, the target node in each node is determined according to the conversion rate, including:

[0087] The node with a conversion rate less than a preset threshold value in each node is determined, and the node with a conversion rate less than a preset threshold value in each node is determined as the target node in each node.

[0088] The funnel principle is used to quantitatively analyze the conversion rate on the user click path. The funnel model analysis is a process data analysis, which can reflect the conversion rate of each node on the click path of the user. In a broad sense, the funnel model can be regarded as a special case of path analysis, which is a path analysis for a few manually specified event nodes. Path analysis usually tracks and records each behavior path of each user, which involves the source and destination of each step, and the conversion rate of each step. Path analysis is exploratory to mine the overall behavior path, find the mainstream path of the user, and even possibly discover some interesting pattern paths that were previously unknown. The funnel model is a priori, manual and proactive to set several key event node paths.

[0089] The execution subject can determine the target node in each node according to the conversion rate through the funnel model. The funnel model refers to finding the user loss link by recording the user conversion rate of different key nodes, and the execution subject can determine the node corresponding to the user loss link as the target node, and the user conversion rate of the target node is low, so as to find the reason for user loss and find the way to improve the conversion rate. The conversion rate refers to the percentage of users who achieve the expected goal (conversion) among all users who have seen the advertisement / website in a statistical period. (Conversion rate = (conversion times / click volume) x 100%).

[0090] The execution subject determines the node with a conversion rate less than a preset threshold at the user loss link as the target node. Of course, the execution subject can also determine all nodes at the user loss link as the target node. The present application embodiment does not make specific limitation on the preset threshold.

[0091] After determining the target node, the execution subject can call the frequent pattern mining program to determine the frequent path according to the related information (such as conversion rate) of the target node. The frequent path can be a path obtained by connecting the nodes with a conversion rate greater than a preset threshold.

[0092] In step S104, the frequent path is input into the prediction model to generate a predicted user click path, and then the website structure corresponding to the website identifier is adjusted based on the predicted user click path.

[0093] The training and evaluation of the prediction model are based on the TensorFlow deep learning computing framework, which is an open-source deep learning tool of Google and covers most techniques used in deep learning, and has very good support for some basic models such as CNN, RNN, LSTM, etc. The training sample set is obtained, the click path data in the obtained training sample set is divided into a training set and a test set, and the prediction model is trained and tested. The prediction model uses a combination of word2vec and LSTM. First, the semantic vector coding of the website URL is obtained using word2vec, and then the obtained URL vector coding is used to train the LSTM model to obtain the prediction model. Using the prediction model, the probability of visiting other web pages under the current state can be predicted to evaluate the rationality of the information structure of the website.

[0094] Based on the trained prediction model, the execution subject can input the frequent path into the prediction model, and the prediction model can output the corresponding probability of each page as the next access page of the input frequent path. For example, when the candidate pages are only page 1, page 2 and page 3, the corresponding probability of page 1 is 20%, the corresponding probability of page 2 is 30%, and the corresponding probability of page 3 is 50%. The execution subject can select the page corresponding to the maximum probability as the next access page of the input frequent path, and sequentially connect the page corresponding to the maximum probability as a vertex into the frequent path to obtain the predicted user click path. And adjust the website structure corresponding to the website identifier based on the predicted user click path. In the embodiments of the present application, the website structure can include website display content, layout of each page of the website and display order of each page. The specific content of the website structure is not limited in the embodiments of the present application.

[0095] The embodiment receives a user behavior processing request, obtains a corresponding website identifier and a user identifier, obtains a corresponding website log based on the website identifier, determines a user click path corresponding to the user identifier based on the website log, obtains a conversion rate of each node in the user click path, determines a target node in each node according to the conversion rate, and then calls a frequent pattern mining program to determine a frequent path according to the target node. The frequent path is input into a prediction model to generate a predicted user click path, and then the website structure corresponding to the website identifier is adjusted based on the predicted user click path. The node with a low single-step conversion rate found by the funnel model is taken as the target node, and the number of clicks of the target node is taken as a threshold value input into the frequent pattern mining program, so that the associated frequent path is mined to the maximum extent. By using the method combining the funnel model and the frequent pattern mining, the conversion rate of the website is analyzed, the big data processing technology is used to quickly mine the massive click path data, the analysis period is shortened, and the user click path is predicted based on the neural network prediction model, which provides a reference basis for website operators to optimize the information structure of key pages, improves the generality and accuracy of the method for processing user behavior, and improves the overall processing efficiency of user behavior analysis.

[0096] Figure 2 is the main flow schematic diagram of the user behavior processing method according to the second embodiment of the application, as shown in Figure 2 the user behavior processing method comprises:

[0097] Step S201, receiving a user behavior processing request, obtaining a corresponding website identifier and a user identifier, and obtaining a corresponding website log based on the website identifier.

[0098] Step S202, determining a user click path corresponding to the user identifier based on the website log, and obtaining a conversion rate of each node in the user click path.

[0099] Specifically, determining a user click path corresponding to the user identifier based on the website log comprises:

[0100] Obtaining a session identifier corresponding to the user identifier in the website log, for example, HH, and the specific representation form of the session identifier is not limited in the embodiment of the application; obtaining corresponding session data based on the session identifier, the execution subject can locate the session identifier, and then obtain the session data corresponding to the session identifier, which can include a webpage link or a webpage jump identifier; and constructing a user click path corresponding to the user identifier based on the session data.

[0101] For example, the execution subject can arrange the session data in chronological order, and then extract the web links or web jump identifiers (for example, A page jumps to B page, and the corresponding web jump identifier can be A-B, and the specific form of the web jump identifier is not limited in the embodiments of the application) in the arranged session data, respectively, determine from which web page to jump to which web page according to the web links or directly determine the web pages jumping to each other according to the web jump identifiers, and then connect the web pages jumping to each other (for example, web page A jumps to web page B) as nodes according to the order of the arranged session data to obtain the user click path, for example, A-B.

[0102] In step S203, the target node in each node is determined according to the conversion rate.

[0103] The target node can be a node with a low conversion rate. Specifically, the node with a conversion rate lower than a preset threshold.

[0104] In step S204, the target node session corresponding to the target node is screened.

[0105] The execution subject can screen the target node session corresponding to the target node in the website log.

[0106] In step S205, the conversion rate of the target node is determined, and the conversion rate of the target node is determined as the support threshold.

[0107] The execution subject obtains the conversion rate of the target node according to the screened target node, and then takes the conversion rate of the target node as the support threshold. For example, the support threshold in the embodiments of the application is: support threshold = number of people converted by a node / number of people corresponding to the click events of the node.

[0108] In step S206, the target node session and the support threshold are input into the frequent pattern mining program to obtain a frequent path.

[0109] The execution subject can input the support threshold as the lowest support threshold set in the frequent pattern mining program, so that the frequent pattern mining program obtains the frequent path according to the support threshold and the input target node session.

[0110] Specifically, the user behavior processing method further includes:

[0111] In response to the frequent path being empty, the support threshold is adjusted based on a preset threshold gradient until the obtained frequent path is non-empty.

[0112] If the returned frequent path is empty, it means that the input threshold is too large, and the execution subject can appropriately reduce the input threshold according to a preset threshold gradient, and re-mine the frequent path. The preset threshold gradient is, for example, 0.5, 0.45, 0.4, 0.35, 0.3. When the support threshold is 0.5, the returned frequent path is empty, the execution subject can take 0.45 as a new support threshold to re-obtain the corresponding frequent path. If the returned frequent path is still empty, the execution subject continues to take 0.4 as a new support threshold, until a non-empty frequent path is obtained, and the obtained frequent path is output.

[0113] Specifically, the frequent path is obtained, including:

[0114] According to the target node session, each candidate path is generated; the support of each candidate path is determined; and the candidate path corresponding to the support greater than the support threshold is determined as the frequent path.

[0115] The support of each candidate path can be the sum or average of the supports of the nodes in the candidate path. The execution subject mines the high-frequency path, i.e., the frequent path, from each candidate path based on the support of each candidate path according to the input support threshold. The execution subject can use the Spark and Hadoop distributed computing framework to accelerate the counting process and improve the performance of the entire calculation.

[0116] An example of the execution process of mining the frequent path includes three steps:

[0117] (1) Screening the session. The session containing the target node is screened out for the next step of frequent path mining, which excludes other sessions and reduces a large amount of operation time.

[0118] (2) Frequent path mining. The frequent path is mined according to the input support threshold. The Spark and Hadoop distributed computing framework is used to accelerate the counting process and improve the performance of the entire calculation. The candidate path support calculation in the frequent path mining algorithm is an I / O intensive calculation. During the entire algorithm execution process, a large part of the time is spent on waiting for I / O. Since the data set itself has a large amount of data, the Spark and Hadoop distributed computing framework is considered to shorten the time consumption caused by I / O, accelerate the AprioriSome algorithm counting process, and thus improve the performance of the entire calculation process.

[0119] (3) Checking the frequent path. If the returned frequent path is empty, it means that the input threshold is too large, and the execution subject can appropriately reduce the input threshold to re-mine the frequent path.

[0120] The frequent path mining design in step (2) uses AprioriSome. The frequent path mining process is split into a control module and a counting module. The control module is responsible for controlling the execution logic, implements the AprioriSome algorithm, calls the counting module to calculate the support of the candidate path, finally obtains the frequent path satisfying the specified support, and returns the frequent path containing the target node specified by the operator.

[0121] In step S207, the frequent path is input into the prediction model to generate a predicted user click path, and then the website structure corresponding to the website identifier is adjusted based on the predicted user click path.

[0122] Specifically, the predicted user click path is generated, including:

[0123] The frequent path is converted into a frequent path vector; the user click distribution is predicted based on the frequent path vector, and then the predicted user click path is generated.

[0124] Specifically, the website structure corresponding to the website identifier is adjusted based on the predicted user click path, including: obtaining the current access path corresponding to the user identifier, determining the access probability of each node in the predicted user click path; determining the node corresponding to the maximum access probability, and then generating the predicted website structure according to the current access path and the node corresponding to the maximum access probability; obtaining the expected website structure, and then updating the expected website structure based on the predicted website structure.

[0125] The jump structure between webpages forms a graph structure, and the webpage is a vertex and the jump of browsing is an edge. The semantic features of each URL page can be described by using the word vector; by using the recurrent neural network LSTM, the probability distribution of the next access page of the current access path can be predicted; the page corresponding to the maximum access probability is determined, and the corresponding page is connected as a vertex into the current access path to form a predicted website structure composed of the predicted user click path. The predicted website structure obtained is compared with the expected website structure (wherein the expected website structure is the originally set website structure), and when there is a difference, the subject can slowly adjust the expected website structure according to the difference to make the expected website structure approach the predicted website structure until they coincide.

[0126] Figure 3 Fig. 1 is a schematic diagram of an application scene of a user behavior processing method according to the third embodiment of the present application. The user behavior processing method of the embodiments of the present application can be applied to a user behavior analysis scene. As shown in Fig. 1, the user behavior processing method of the embodiments of the present application can be applied to a user behavior analysis scene. Figure 3As shown, the server 302 receives a user behavior processing request 301, obtains a corresponding website identifier 303 and a user identifier 304, and obtains a corresponding website log 305 based on the website identifier 303. The server 302 determines a user click path 306 corresponding to the user identifier 304 based on the website log 305, and obtains a conversion rate 307 of each node in the user click path 306. The server 302 determines a target node 308 in each node according to the conversion rate 307, and then calls a frequent pattern mining program to determine a frequent path 309 according to the target node 308. The server 302 inputs the frequent path 309 to a prediction model 310 to generate a predicted user click path 311, and then adjusts a website structure 312 corresponding to the website identifier 303 based on the predicted user click path 311.

[0127] As an embodiment of the present application, for example, the browsing and clicking behavior of a user of a Web site is recorded in a Web site server log. By mining the Web site log, an e-commerce enterprise can more clearly understand the browsing interests and behavior patterns of the user. This helps the enterprise to more accurately and effectively formulate a marketing strategy, improve the conversion rate of the website, and also optimize the information structure of the website. The present application provides a user behavior analysis system (which can be an execution subject) for a large e-commerce website, which is used to expand the application range of Web log mining and further play the use value and significance of Web log mining. By providing a system including a click path construction function, a conversion rate analysis function, and an information structure evaluation function, the user's click path is constructed from the website log, and a method combining a funnel model and frequent pattern mining is used to analyze the conversion rate of the website. A big data processing technology is used to quickly model and mine these massive click paths, thereby shortening the analysis period. A neural network is used to predict the user clicks, thereby providing a reference basis for website operators to optimize the information structure of key pages.

[0128] The user behavior analysis system of the embodiment of the present application is composed of three core modules, namely a click path construction module, a conversion rate analysis module, and an information structure evaluation module. The functions from processing of original website logs to analysis and display of the conversion rate and information structure of the website are realized. The user behavior analysis system of the embodiment of the present application performs the following logic:

[0129] The method for constructing the user click path and the scalability are solved. After the steps of Web site log cleaning, user identification, session identification, and user click path restoration, the Web site log is processed, and the scalability problem of massive data processing is solved, i.e., the user behavior analysis system can be horizontally expanded at any time to cope with the rapid growth of data. In the process of Web log preprocessing, a MapReduce distributed computing framework is used to process a large amount of unstructured Web log data.

[0130] The funnel model combines frequent pattern mining to analyze the conversion rate of the website. The conversion rate of the problem node is taken as the input of the frequent pattern mining by using the funnel model to analyze the conversion rate of the page, and the reason for the low conversion rate in the user click path is analyzed to provide a reference basis for the website operator to improve the conversion rate of the website. Meanwhile, in the process of analyzing the conversion rate, the characteristics of Hadoop and Spark are fully utilized, and the two are combined for use, the idea of distributed computing is fully utilized, the large log file is split, and the efficiency problem of the overall solution is improved by using the parallel computing of Spark. On the one hand, the scalability problem of massive data calculation is solved, and on the other hand, the performance loss caused by Hadoop is solved, and the overall analysis efficiency is improved.

[0131] The rationality of the website information structure is analyzed based on neural network. The technical knowledge in the field of artificial intelligence is introduced, the neural network technology is used to replace the traditional feature extraction analysis to predict the algorithm, the generality and expansibility of the algorithm are improved, the user click path is vectorized through word2vec, and then the neural network LSTM model is trained and used to predict the user click distribution. Through analyzing the prediction result, the correlation between the website link and the user click behavior is analyzed to guide the website operator to optimize the website information structure.

[0132] The original website log is processed by using the MapReduce programming model, the real click path of the user is restored, and the basic data is provided for the conversion rate analysis and website information structure evaluation. On the basis of the user click path data, the funnel model and the frequent path mining are combined to analyze the conversion rate of the website user. That is, the click number of the node with low single-step conversion rate found by the funnel model is taken as the threshold input of the frequent pattern mining program, so that the associated frequent path is mined to the maximum extent, and a basis is provided for the e-commerce website operator to optimize the website conversion rate. The word vector tool word2vec and the recurrent neural network LSTM are used to learn the user click path data, so as to obtain a prediction model, predict the user click behavior, help the website operator to find out whether the page information structure is reasonable, and provide a reference basis for the information structure optimization of the website. In engineering implementation, the method used in the embodiment of the application greatly improves the performance of the traditional method, and solves the scalability problem of massive data analysis of the large e-commerce website.

[0133] Figure 4 is a schematic diagram of the main unit of the user behavior processing device according to the embodiment of the application. As shown in Figure 4 , the user behavior processing device includes a receiving unit 401, an obtaining unit 402, a frequent path determination unit 403, and an adjusting unit 404.

[0134] The receiving unit 401 is configured to receive a user behavior processing request, acquire a corresponding website identifier and a user identifier, and acquire a corresponding website log based on the website identifier;

[0135] The acquiring unit 402 is configured to determine a user click path corresponding to the user identifier based on the website log, and acquire a conversion rate of each node in the user click path.

[0136] The frequent path determining unit 403 is configured to determine a target node in each node according to the conversion rate, and then call a frequent pattern mining program to determine a frequent path according to the target node.

[0137] The adjusting unit 404 is configured to input the frequent path into a prediction model to generate a predicted user click path, and then adjust a website structure corresponding to the website identifier based on the predicted user click path.

[0138] In some embodiments, the frequent path determining unit 403 is further configured to determine a node with a conversion rate less than a preset threshold value in each node, and determine the node with the conversion rate less than the preset threshold value as the target node in each node.

[0139] In some embodiments, the frequent path determining unit 403 is further configured to filter a target node session corresponding to the target node, determine a conversion rate of the target node, determine the conversion rate of the target node as a support threshold value, and input the target node session and the support threshold value into the frequent pattern mining program to obtain the frequent path.

[0140] In some embodiments, the adjusting unit 404 is further configured to adjust the support threshold value based on a preset threshold gradient in response to the frequent path being empty, until the obtained frequent path is non-empty.

[0141] In some embodiments, the frequent path determining unit 403 is further configured to generate each candidate path according to the target node session, determine a support of each candidate path, and determine a candidate path corresponding to a support greater than the support threshold value as the frequent path.

[0142] In some embodiments, the adjusting unit 404 is further configured to convert the frequent path into a frequent path vector, predict a user click distribution based on the frequent path vector, and then generate the predicted user click path.

[0143] In some embodiments, the acquiring unit 402 is further configured to acquire a session identifier corresponding to the user identifier in the website log, acquire corresponding session data based on the session identifier, and construct a user click path corresponding to the user identifier based on the session data.

[0144] In some embodiments, the adjusting unit 404 is further configured to: acquire a current access path corresponding to the user identifier, determine an access probability of each node in the predicted user click path; determine a node corresponding to a maximum access probability, and then generate a predicted website structure according to the current access path and the node corresponding to the maximum access probability; and acquire an expected website structure, and then update the expected website structure based on the predicted website structure.

[0145] It should be noted that the user behavior processing method and the user behavior processing apparatus of the present application have a corresponding relationship in the specific implementation content, and therefore repeated content will not be described.

[0146] Figure 5 An exemplary system architecture 500 to which the user behavior processing method or the user behavior processing apparatus of the embodiments of the present application can be applied is shown.

[0147] As shown in Figure 5 , the system architecture 500 can include terminal devices 501, 502, 503, a network 504, and a server 505. The network 504 serves as a medium to provide communication links between the terminal devices 501, 502, 503 and the server 505. The network 504 can include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0148] A user can use the terminal devices 501, 502, 503 to interact with the server 505 through the network 504 to receive or send messages, etc. Various communication client applications can be installed on the terminal devices 501, 502, 503, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0149] The terminal devices 501, 502, 503 can be various electronic devices with user behavior processing screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers, etc.

[0150] The server 505 can be a server providing various services, for example, a background management server (only as an example) providing support for a user behavior processing request submitted by a user using a terminal device 501, 502, 503. The background management server can receive a user behavior processing request, obtain a corresponding website identifier and a user identifier, obtain a corresponding website log based on the website identifier, determine a user click path corresponding to the user identifier based on the website log, obtain a conversion rate of each node in the user click path, determine a target node in each node according to the conversion rate, and then call a frequent pattern mining program to determine a frequent path according to the target node; input the frequent path into a prediction model to generate a predicted user click path, and then adjust the website structure corresponding to the website identifier based on the predicted user click path. The node with a low single-step conversion rate found by the funnel model is taken as the target node, and the number of clicks of the target node is taken as a threshold input to the frequent pattern mining program, so as to maximize the mining of the associated frequent path. By using the method combining the funnel model and the frequent pattern mining, the conversion rate of the website is analyzed, the big data processing technology is used to quickly mine these massive click path data, the analysis period is shortened, and the prediction model based on the neural network is used to predict the user click path, so as to provide a reference basis for website operators to optimize the information structure of the key page, improve the generality and accuracy of the method for processing the user behavior, and improve the overall processing efficiency of the user behavior analysis.

[0151] It should be noted that the user behavior processing method provided by the embodiments of the present application is generally executed by the server 505, and correspondingly, the user behavior processing apparatus is generally arranged in the server 505.

[0152] It should be understood that Figure 5 The number of terminal devices, networks and servers in the above-mentioned system is only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.

[0153] Reference will be made to Figure 6 which shows a structural schematic diagram of a computer system 600 of a terminal device suitable for implementing the embodiments of the present application. Figure 6 The terminal device shown is only an example and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0154] As Figure 6As shown, the computer system 600 includes a central processing unit (CPU) 601 which can perform various appropriate actions and processes according to programs stored in a read only memory (ROM) 602 or loaded into a random access memory (RAM) 603 from a storage section 608. In the RAM 603, various programs and data required for the operation of the computer system 600 are also stored. The CPU 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0155] Connected to the I / O interface 605 are an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a display device such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as necessary. A removable recording medium 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 610 as necessary, so that a computer program read therefrom is installed into the storage section 608 as necessary.

[0156] In particular, according to the embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments of the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program codes for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication section 609, and / or installed from the removable recording medium 611. When the computer program is executed by the central processing unit (CPU) 601, the above-described functions defined in the system of the present application are executed.

[0157] It should be noted that computer-readable media in this disclosure can be computer-readable storage media, or computer-readable signal media, or any combination thereof. Computer-readable storage media can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the above. More specific examples of computer-readable storage media can 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 above. In this disclosure, computer-readable storage media can 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 this disclosure, a computer-readable signal medium can include a computer-readable storage medium, or any computer-readable medium that can transmit or propagate code in the form of computer-readable instructions or program code, or any combination of the above. The computer-readable medium can be transmitted in baseband or as part of a carrier wave over a transmission medium, including a wired medium, or a wireless medium, or any suitable combination of the above. Computer-readable media can also be any medium that can be used to store or transfer a program for use by or in connection with an instruction execution system, apparatus, or device.

[0158] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functional processes, and operations that can be implemented in systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flow diagrams or block diagrams can represent a module, a segment, or a portion of code that 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 can 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 can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flow diagrams, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

[0159] The units described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. The described units can also be arranged in a processor, for example, a processor can be described as including a receiving unit, an obtaining unit, a frequent path determining unit and an adjusting unit. In some cases, the names of these units do not constitute a limitation on the units themselves.

[0160] As another aspect, the present application also provides a computer readable medium, which can be included in the device described in the above embodiments, or can exist independently without being assembled into the device. The computer readable medium carries one or more programs, which, when executed by the device, cause the device to receive a user behavior processing request, obtain a corresponding website identifier and a user identifier, obtain a corresponding website log based on the website identifier, determine a user click path corresponding to the user identifier based on the website log, obtain a conversion rate of each node in the user click path, determine a target node in each node according to the conversion rate, and then call a frequent pattern mining program to determine a frequent path according to the target node, input the frequent path into a prediction model to generate a predicted user click path, and then adjust a website structure corresponding to the website identifier based on the predicted user click path.

[0161] The computer program product of the present application includes a computer program, which, when executed by a processor, implements the user behavior processing method in the embodiments of the present application.

[0162] According to the technical solution of the embodiments of the present application, the node with a low single-step conversion rate found by the funnel model is taken as a target node, and the number of clicks of the target node is taken as a threshold input to the frequent pattern mining program, so as to maximize the mining of associated frequent paths. By using the method combining the funnel model and the frequent pattern mining, the conversion rate of the website is analyzed, the big data processing technology is used to quickly mine these massive click path data, the analysis period is shortened, and the prediction model based on the neural network is used to predict the user click path, so as to provide a reference basis for website operators to optimize the information structure of key pages, improve the generality and accuracy of the method for processing user behavior, and improve the overall processing efficiency of user behavior analysis.

[0163] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made depending on design requirements and other factors. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A user behavior processing method, characterized in that, include: Receive user behavior processing requests, obtain the corresponding website identifier and user identifier, and obtain the corresponding website logs based on the website identifier; Based on the website logs, session data is obtained, the user click path corresponding to the user identifier is determined, and the conversion rate of each node in the user click path is obtained. Based on the conversion rate, the node corresponding to the user churn stage is determined as the target node among the nodes. Then, the Spark and Hadoop distributed computing framework is used to call the frequent pattern mining program to determine the frequent path based on the target node. This includes: the conversion rate of the target node is the support threshold, the target node session and the support threshold are input into the frequent pattern mining program to obtain the frequent path; wherein, in response to the frequent path being empty, the support threshold is adjusted based on a preset threshold gradient until the obtained frequent path is not empty. The frequent paths are input into the prediction model to generate predicted user click paths, including: converting the frequent paths into frequent path vectors; predicting user click distribution based on the frequent path vectors, and then generating predicted user click paths. Adjusting the website structure corresponding to the website identifier based on the predicted user click path includes: obtaining the current access path corresponding to the user identifier, determining the access probability of each node in the predicted user click path; determining the node corresponding to the maximum access probability, and then generating a predicted website structure based on the current access path and the node corresponding to the maximum access probability; obtaining the expected website structure, and then updating the expected website structure based on the predicted website structure.

2. The method according to claim 1, characterized in that, Determining the target node among the nodes based on the conversion rate includes: Identify nodes whose conversion rate is less than a preset threshold. The nodes whose conversion rate is less than a preset threshold are identified as target nodes among the nodes.

3. The method according to claim 1, characterized in that, The frequent pattern mining program, used to determine frequent paths based on the target node, includes: The target node sessions corresponding to the target node are obtained through filtering. Determine the conversion rate of the target node, and use the conversion rate of the target node as the support threshold; The target node session and the support threshold are input into the frequent pattern mining program to obtain frequent paths.

4. The method according to claim 3, characterized in that, The obtained frequent paths include: Each candidate path is generated based on the target node session; Determine the support level for each candidate path; Candidate paths with support values ​​greater than the support threshold are identified as frequent paths.

5. The method according to claim 1, characterized in that, Determining the user click path corresponding to the user identifier based on the website logs includes: Obtain the session identifier corresponding to the user identifier from the website logs; Obtain the corresponding session data based on the session identifier; The user click path corresponding to the user identifier is constructed based on the session data.

6. A user behavior processing device, characterized in that, include: The receiving unit is configured to receive user behavior processing requests, obtain the corresponding website identifier and user identifier, and obtain the corresponding website logs based on the website identifier. The acquisition unit is configured to obtain session data based on the website logs, determine the user click path corresponding to the user identifier, and obtain the conversion rate of each node in the user click path; The frequent path determination unit is configured to determine the node corresponding to the user churn link based on the conversion rate as the target node among the nodes, and then use the Spark and Hadoop distributed computing framework to call the frequent pattern mining program to determine the frequent path based on the target node. An adjustment unit is configured to input the frequent paths into a prediction model to generate predicted user click paths, including: converting the frequent paths into frequent path vectors; predicting user click distributions based on the frequent path vectors, thereby generating predicted user click paths; and adjusting the website structure corresponding to the website identifier based on the predicted user click paths, including: obtaining the current access path corresponding to the user identifier and determining the access probability of each node in the predicted user click path; determining the node corresponding to the maximum access probability, and then generating a predicted website structure based on the current access path and the node corresponding to the maximum access probability; obtaining the expected website structure, and then updating the expected website structure based on the predicted website structure. The adjustment unit is further configured to: set the conversion rate of the target node as the support threshold, input the target node session and the support threshold into a frequent pattern mining program to obtain frequent paths; wherein, in response to the frequent paths being empty, adjust the support threshold based on a preset threshold gradient until the obtained frequent paths are not empty.

7. The apparatus according to claim 6, characterized in that, The frequent path determination unit is further configured to: Identify nodes whose conversion rate is less than a preset threshold. The nodes whose conversion rate is less than a preset threshold are identified as target nodes among the nodes.

8. The apparatus according to claim 7, characterized in that, The frequent path determination unit is further configured to: The target node sessions corresponding to the target node are obtained through filtering. Determine the conversion rate of the target node, and use the conversion rate of the target node as the support threshold; The target node session and the support threshold are input into the frequent pattern mining program to obtain frequent paths.

9. The apparatus according to claim 8, characterized in that, The frequent path determination unit is further configured to: Each candidate path is generated based on the target node session; Determine the support level for each candidate path; Candidate paths with support values ​​greater than the support threshold are identified as frequent paths.

10. A user behavior processing electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-5.

11. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.

Images