Push solution optimization method and system based on user-side response data collection

Through user behavior feature extraction and distribution model identification, user-side data collection is optimized, the problem of messy user behavior data is solved, and the accuracy and user experience of push solutions are improved.

CN118797175BActive Publication Date: 2024-12-06SHENZHEN SINOAGRI E-COMMERCE CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411281697.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-12-06
Estimated Expiration
2044-09-13

AI Technical Summary

Technical Problem

The user behavior data collected by the user side is too messy, resulting in a shift in the push scheme and a decline in user experience.

Method used

By obtaining historical user behavior data, using the pre-trained user browsing behavior-interest recognition model for feature extraction and classification, building user hobbies and browsing behavior habit portraits, combining the Dirichlet distribution algorithm to build the behavior-interest-vocabulary distribution model, and identifying effective Information vocabulary, optimize information push.

Benefits of technology

It improves the accuracy of the optimization direction of the push solution, reduces noise data, and improves the user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118797175B_ABST
    Figure CN118797175B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of big data push technology, a push scheme optimization method and system based on user-side response data collection, including: feature recognition of historical user behavior data sets to obtain various interest tags, weight scores and behavior classification tags; constructing user hobby portraits and browsing behavior habit portraits based on interest tags, weight scores and behavior classification tags; classifying real-time behavior data based on browsing behavior habit portraits to obtain behavior classification tags; weighting real-time behavior data according to weight score configuration rules and behavior classification tags to obtain weighted behavior data; constructing a behavior-lexicon distribution model based on the Dirichlet distribution algorithm and the joint distribution algorithm to identify valid information vocabulary in the weighted behavior data; optimizing the user hobby portrait based on the valid information vocabulary to obtain an optimized information push result. The present invention can solve the problem that the user behavior data collected by the user end is too messy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of big data push technology, and in particular to a push scheme optimization method and system based on user-side response data collection. Background Art

[0002] With the development of informatization, not only users actively look for information and products, but information and products also actively look for users, thus big data push services come into being. Nowadays, this kind of data push exists in all aspects of life, such as entertainment, news and shopping. The data push refers to predicting the interests and hobbies of users based on their behavior information, and then pushing relevant content to them to increase the amount of information reading or product purchases.

[0003] Nowadays, most software with data push services have huge amounts of data and search functions. For example, if you open a website about agriculture, you can get some of the latest and important agricultural information, but it may not be the information that users need to know. Users can search for the content they need on the website. At this time, the website collects the user's search target and pushes the search target and related information to the customer.

[0004] However, life is not scripted. When users operate the client, their behavior is relatively random, resulting in large noise and redundancy in the data collected by the user end. In addition, users are also disturbed by real life and retrieve or browse some information that they do not use frequently, which causes the push plan to be offset and the user experience to deteriorate. Summary of the invention

[0005] The present invention provides a method and system for optimizing a push scheme based on user-side response data collection, the main purpose of which is to solve the problem that the user behavior data collected by the user side is too messy.

[0006] To achieve the above-mentioned purpose, the present invention provides a method for optimizing a push scheme based on user terminal response data collection, comprising:

[0007] Obtain a set of historical user behavior data within a preset time period;

[0008] Using a pre-trained user browsing behavior-interest recognition model, the historical user behavior data set is quantized and encoded to obtain a historical behavior feature vector set, and each historical behavior feature vector in the historical behavior feature vector set is concatenated with a vector of a preset character length to obtain a historical behavior feature matrix;

[0009] Using the pre-constructed convolution kernel set in the user browsing behavior-interest recognition model, a convolution operation is performed on the historical behavior feature matrix to obtain a convolution matrix set;

[0010] Using the pre-constructed pooling network in the user browsing behavior-interest recognition model, a maximum pooling operation is performed on the convolution matrix set to obtain a pooling matrix set;

[0011] Using the pre-constructed flattened network in the user browsing behavior-interest recognition model, the pooling matrix set is subjected to feature concatenation and dimensionality reduction operations to obtain a historical behavior feature sequence;

[0012] Using the fully connected network pre-constructed in the user browsing behavior-interest recognition model, the historical behavior feature sequence is subjected to a binary classification operation based on user behavior and behavior object, thereby obtaining a user browsing behavior feature sequence and a user interest feature sequence respectively;

[0013] Performing a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag, and performing a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification tag corresponding to each historical user behavior data in the historical user behavior data set, wherein the behavior classification tag includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type;

[0014] According to the interest tags and weight scores, a user hobby portrait is constructed, and a display information set pushed by a pre-constructed data push service according to the user hobby portrait is obtained, and a browsing behavior habit portrait is constructed according to the behavior classification tags corresponding to each of the historical user behavior data;

[0015] Acquire real-time behavior data of the user's response to the display information set, and identify behavior classification labels of the real-time behavior data according to the browsing behavior habit portrait;

[0016] When the behavior classification label of the real-time behavior data is the false touch type, determining that the real-time behavior data is invalid;

[0017] When the behavior classification label of the real-time behavior data is a behavior classification label other than the false touch type, weighting the real-time behavior data according to a pre-constructed weight score configuration rule to obtain weighted behavior data;

[0018] According to the pre-constructed Dirichlet distribution algorithm, a behavior-interest distribution model and an interest-lexicon distribution model are constructed for the historical user behavior data set, and the joint distribution of the behavior-interest distribution model and the interest-lexicon distribution model is calculated to obtain a behavior-lexicon distribution model. The behavior-lexicon distribution model is used to identify valid information words in the weighted behavior data, wherein the behavior-interest distribution model is expressed as:

[0019]

[0020]

[0021] In the formula, Represents historical user behavior data Distribution of interests, Indicates Historical user behavior data, represents the Dirichlet distribution, represents the total number of historical user behavior data in the historical user behavior data set, represents the prior parameters of the Dirichlet distribution; Indicates the generation of interest tags in the entire historical user behavior data set The probability of Indicates interest tags, represents the Laplace operator, Indicates user Generate interest tags based on historical user behavior data The probability of Represents historical user behavior data;

[0022] The interest-vocabulary distribution model is expressed as:

[0023]

[0024]

[0025] In the formula, Indicate interest tag The vocabulary distribution of Indicates interest tags, represents the total number of interest tags in the historical user behavior data set, represents the prior parameters of the Dirichlet distribution, Represents vocabulary, Represents the vocabulary in the historical user behavior data set Probability of generation;

[0026] The behavior-vocabulary distribution model is expressed as:

[0027]

[0028] In the formula, It means that the given prior probability is and Time, vocabulary and interests Distribution of

[0029] The valid information vocabulary is sent to the data push service, the user preference portrait is optimized, and an optimized information push result is obtained.

[0030] Optionally, obtaining a set of historical user behavior data within a preset time period includes:

[0031] Integrate historical user behavior data within a preset time period from multiple data sources to obtain an initial data set;

[0032] Performing a duplicate check on the initial data set, deleting duplicate behavior data in the initial data set to obtain a clean data set, and performing data standardization on the clean data set to obtain a standard data set, wherein the data standardization includes lowercase conversion, stop word removal, and Z-score standardization;

[0033] The TF-IDF index of each word in the standard data set is calculated, and important features of each word in the standard data set are extracted according to the TF-IDF index to obtain a historical user behavior data set.

[0034] Optionally, performing a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag includes:

[0035] Performing feature recognition on the user interest feature sequence to obtain an interest tag set;

[0036] Marking each interest tag in the interest tag set on the corresponding historical user behavior data, and equipping each interest tag with statistical parameters;

[0037] Determine whether each of the interest tags is marked for the first time in the corresponding historical user behavior data;

[0038] When the interest tag is marked for the first time, adding one to the statistical parameter of the interest tag;

[0039] When the interest tag is not marked for the first time, a zero addition operation is performed on the statistical parameter of the interest tag;

[0040] Statistical parameters corresponding to each interest tag are obtained, and each of the statistical parameters is normalized to obtain a weight score corresponding to each of the interest tags.

[0041] Optionally, the performing a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification label corresponding to each historical user behavior data in the historical user behavior data set includes:

[0042] Performing behavior recognition on the user browsing behavior feature sequence to obtain a user behavior recognition result set, and performing a normalization operation on the user behavior recognition result set to obtain a normalized behavior recognition result set;

[0043] According to a pre-built K-means clustering algorithm, a grouping clustering operation based on four clusters is performed on the normalized behavior recognition result set to obtain a behavior summary cluster set;

[0044] Calculating the cluster center position of each behavior summary cluster in the behavior summary cluster set, and arranging each behavior summary cluster according to each cluster center position to obtain a position arrangement result;

[0045] According to the position arrangement result, a behavior classification label is sequentially configured for each of the behavior summary clusters, wherein the behavior classification label includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type.

[0046] Optionally, identifying a behavior classification label of the real-time behavior data according to the browsing behavior habit portrait includes:

[0047] Performing transfer learning on the pre-built user behavior classification model according to some network parameters in the user browsing behavior-interest recognition model to obtain an updated user behavior classification model;

[0048] Fine-tune the updated user behavior classification model according to the browsing behavior habit portrait to obtain a local user behavior classification model;

[0049] Using the local user behavior classification model, extracting features from the real-time behavior data to obtain a real-time behavior feature sequence;

[0050] Performing spatial mapping on the real-time behavior feature sequence to obtain the real-time behavior position;

[0051] According to the center positions of each cluster, the real-time behavior position is identified as belonging to a cluster, the belonging behavior summary cluster of the real-time behavior data is obtained, and the behavior classification label corresponding to the belonging behavior summary cluster is obtained.

[0052] Optionally, the using the behavior-vocabulary distribution model to identify valid information vocabulary in the weighted behavior data includes:

[0053] The behavior-vocabulary distribution model is used to calculate the interest latent content of the weighted behavior data, and the interest latent content is expressed as:

[0054]

[0055] In the formula, Indicates the weighted behavior data The first The implicit interest content corresponding to each word is Representing interest distribution The multinomial distribution of represents the total number of user behaviors in the weighted behavior data, represents the total number of words in the weighted behavioral data;

[0056] The acquisition amount of each word of the weighted behavior data is calculated according to the interest implicit content, and the acquisition amount is expressed as:

[0057]

[0058] In the formula, Indicates the weighted behavior data The first The corresponding acquisition amount of words, Implicit content of interest Multinomial distribution of ;

[0059] According to each of the acquisition amounts, words whose acquisition amounts are greater than a preset qualified threshold are screened to obtain valid information words in the weighted behavior data.

[0060] To achieve the above object, the present invention also provides a push scheme optimization system based on user terminal response data collection, comprising:

[0061] A feature extraction module is used to obtain a historical user behavior data set within a preset time period, obtain a historical user behavior data set within a preset time period, and use a pre-trained user browsing behavior-interest recognition model to perform a feature extraction operation based on user behavior and behavior objects on the historical user behavior data set to obtain a user browsing behavior feature sequence and a user interest feature sequence;

[0062] A feature classification module is used to perform a quantitative statistical operation on the user interest feature sequence based on interest tags to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag, and to perform a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification label corresponding to each historical user behavior data in the historical user behavior data set, wherein the behavior classification label includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type;

[0063] A dual-profile construction module is used to construct a user hobby portrait based on the interest tags and weight scores, obtain a display information set pushed by a pre-built data push service based on the user hobby portrait, and construct a browsing behavior habit portrait based on the behavior classification tags corresponding to each of the historical user behavior data;

[0064] An effective user data collection module is used to obtain the real-time behavior data of the user's response to the display information set, and identify the behavior classification label of the real-time behavior data according to the browsing behavior habit portrait, and when the behavior classification label of the real-time behavior data is the mis-touch type, determine that the real-time behavior data is invalid, and when the behavior classification label of the real-time behavior data is other behavior classification labels other than the mis-touch type, perform weight configuration on the real-time behavior data according to a pre-constructed weight score configuration rule to obtain weighted behavior data, and construct a behavior-interest distribution model and an interest-lexicon distribution model for the historical user behavior data set according to a pre-constructed Dirichlet distribution algorithm, calculate the joint distribution of the behavior-interest distribution model and the interest-lexicon distribution model to obtain a behavior-lexicon distribution model, and use the behavior-lexicon distribution model to identify effective information vocabulary in the weighted behavior data;

[0065] The push optimization module is used to send the valid information vocabulary to the data push service, optimize the user preference portrait, and obtain an optimized information push result.

[0066] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0067] a memory storing at least one instruction; and

[0068] The processor executes the instructions stored in the memory to implement the above-mentioned push solution optimization method based on user terminal response data collection.

[0069] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned push solution optimization method based on user-side response data collection.

[0070] The present invention is to solve the problem described in the background technology. The present invention first extracts and classifies the historical user behavior data set based on user behavior and behavior objects through feature extraction. The user behavior refers to the actions of the user operation, such as reading time, reading speed, reading test, etc., and the behavior object is information such as video and news. These behavior objects will contain corresponding content tags, such as wheat, corn, breeding and pesticides. The present invention can distinguish whether the user is satisfied with the information by analyzing the user behavior, and the analysis of the behavior object can distinguish the user's preference type. The present invention uses a double portrait to set a portrait on the user end and the push end. The browsing behavior habit portrait of the user end is used to collect information recognized by the user, and the push end is used to calculate the user's preferences for push. In addition, when collecting user behavior data, the present invention also further improves the effective information in the user collected data by constructing a behavior-vocabulary distribution model, thereby improving the accuracy of the optimization direction of the push scheme. Therefore, the present invention can solve the problem that the user behavior data collected by the user end is too messy. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A flow chart of a method for optimizing a push solution based on user-side response data collection provided by an embodiment of the present invention;

[0072] Figure 2 A functional module diagram of a push solution optimization system based on user-side response data collection provided by an embodiment of the present invention;

[0073] Figure 3 A schematic diagram of the structure of an electronic device for implementing the push solution optimization method based on user terminal response data collection provided by an embodiment of the present invention.

[0074] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0075] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0076] The embodiment of the present application provides a method for optimizing a push scheme based on user-side response data collection. The execution subject of the method for optimizing a push scheme based on user-side response data collection includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for optimizing a push scheme based on user-side response data collection can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0077] Reference Figure 1 FIG. 1 is a flow chart of a method for optimizing a push scheme based on user terminal response data collection provided by an embodiment of the present invention. In this embodiment, the method for optimizing a push scheme based on user terminal response data collection includes:

[0078] S1. Obtain a historical user behavior data set within a preset time period, and use a pre-trained user browsing behavior-interest recognition model to perform a feature extraction operation based on user behavior and behavior objects on the historical user behavior data set to obtain a user browsing behavior feature sequence and a user interest feature sequence.

[0079] In the embodiment of the present invention, push service updates can be performed according to a preset time frequency, such as once a month, so as to obtain a historical user behavior data set within a month. In the embodiment of the present invention, the historical user behavior data in the historical user behavior data set includes user behavior and behavior objects, wherein the user behavior refers to the actions of user operations, such as reading time, reading speed, and reading test, etc., and the behavior object refers to information such as videos and news. Most of these behavior objects will contain corresponding content tags, such as wheat, corn, breeding, and pesticides. In addition, due to the pre-processing storage process, each historical user behavior data in the default historical user behavior data set represents a behavior.

[0080] In detail, in the embodiment of the present invention, the step of obtaining a set of historical user behavior data within a preset time period includes:

[0081] Integrate historical user behavior data within a preset time period from multiple data sources to obtain an initial data set;

[0082] Performing a duplicate check on the initial data set, deleting duplicate behavior data in the initial data set to obtain a clean data set, and performing data standardization on the clean data set to obtain a standard data set, wherein the data standardization includes lowercase conversion, stop word removal, and Z-score standardization;

[0083] The TF-IDF index of each word in the standard data set is calculated, and important features of each word in the standard data set are extracted according to the TF-IDF index to obtain a historical user behavior data set.

[0084] The multiple data sources refer to the integration of historical user behavior data of the same account on different clients, such as mobile phones, tablets, and computers, to obtain the initial data set. In addition, the same data source may also include web browsing history, purchase records, search records, and comment data.

[0085] Specifically, in the embodiments of the present invention, an initial data set is obtained. Duplicate checking can be performed according to the unique identifiers in the initial data set, and duplicate behavior data is deleted to obtain a clean data set. Then, data cleaning operations are sequentially performed according to lowercasing, stop word removal, and Z-score standardization to obtain a standard data set.

[0086] Among them, the lowercasing is to convert all letters in the clean data set into lowercase letters, which can avoid duplicate or different word counts caused by different letter cases; the stop word removal refers to words that often appear in the clean data set but have no practical significance for text analysis, such as "的", "是", "在", etc. In English, common stop words include "the", "is", "in", etc.; the Z-score standardization is a data standardization technique used to convert data into a distribution with a mean of 0 and a standard deviation of 1.

[0087] Furthermore, the TF-IDF metric is a statistical method for information retrieval and text mining, used to evaluate the importance of a word in a document or a set of documents. It measures the weight of a word by considering the frequency of the word in a specific document and its rarity in the entire document set. In the embodiments of the present invention, by calculating the TF-IDF metric and according to a threshold, important words are screened out to obtain a historical user behavior data set.

[0088] Furthermore, the user browsing behavior-interest recognition model is a neural network model based on text recognition, used to identify the user behaviors and behavior objects in the historical user behavior data set.

[0089] Specifically, in the embodiments of the present invention, the pre-trained user browsing behavior-interest recognition model is used to perform feature extraction operations on the historical user behavior data set based on user behaviors and behavior objects, to obtain a user browsing behavior feature sequence and a user interest feature sequence, including:

[0090] Using the pre-trained user browsing behavior-interest recognition model, perform quantization encoding operations on the historical user behavior data set to obtain a set of historical behavior feature vectors, and perform vector splicing operations of a preset character length on each historical behavior feature vector in the set of historical behavior feature vectors to obtain a historical behavior feature matrix;

[0091] Using the pre-constructed set of convolutional kernels in the user browsing behavior-interest recognition model, perform convolutional operations on the historical behavior feature matrix to obtain a set of convolutional matrices;

[0092] Using the pre-constructed pooling network in the user browsing behavior-interest recognition model, perform max pooling operations on the set of convolutional matrices to obtain a set of pooling matrices;

[0093] Using the pre-constructed flattened network in the user browsing behavior-interest recognition model, the pooling matrix set is subjected to feature concatenation and dimensionality reduction operations to obtain a historical behavior feature sequence;

[0094] By utilizing the fully connected network pre-constructed in the user browsing behavior-interest recognition model, a binary classification operation based on user behavior and behavior object is performed on the historical behavior feature sequence to obtain a user browsing behavior feature sequence and a user interest feature sequence, respectively.

[0095] In an embodiment of the present invention, each word in the historical user behavior data set is converted into a quantized code through a pre-constructed Bag of Words (BoW) model to obtain a set of historical behavior feature vectors, and then according to a preset character length, such as 128 bytes, feature vectors shorter than the character length can be padded with 0, and feature vectors longer than the character length can be truncated, and the rest are padded to the next row, thereby obtaining a historical behavior feature matrix. Then, according to the convolution pooling and flattening operation in the user browsing behavior-interest identification model, a historical behavior feature sequence is obtained. Among them, the convolution pooling and flattening is a common process in convolutional neural networks, which is used for feature extraction and dimensionality reduction, and will not be repeated here. Among them, the bag of words model is a commonly used text representation method, which is widely used in fields such as natural language processing and information retrieval, and is used to represent text (such as sentences, paragraphs or documents) as a "bag of words", that is, only focusing on the frequency of occurrence of words, while ignoring the order and grammatical structure between words.

[0096] Then, in the embodiment of the present invention, the fully connected network pre-constructed in the user browsing behavior-interest recognition model is used to perform a binary classification operation based on user behavior and behavior object, and obtain a user browsing behavior feature sequence and a user interest feature sequence respectively. The training process of the fully connected network can be completed by pre-constructing labeled samples of manually classified behaviors and objects.

[0097] S2. Perform a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain the interest tags that the user is interested in and the weight scores corresponding to the interest tags, and perform a behavior classification operation on the user browsing behavior feature sequence to obtain the behavior classification tags corresponding to the historical user behavior data in the historical user behavior data set, wherein the behavior classification tags include accidental touch type, rough browsing type, ordinary browsing type and careful browsing type.

[0098] In detail, in the embodiment of the present invention, the performing of a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag includes:

[0099] Performing feature recognition on the user interest feature sequence to obtain an interest tag set;

[0100] Marking each interest tag in the interest tag set on the corresponding historical user behavior data, and equipping each interest tag with statistical parameters;

[0101] Determine whether each of the interest tags is marked for the first time in the corresponding historical user behavior data;

[0102] When the interest tag is marked for the first time, adding one to the statistical parameter of the interest tag;

[0103] When the interest tag is not marked for the first time, a zero addition operation is performed on the statistical parameter of the interest tag;

[0104] Statistical parameters corresponding to each interest tag are obtained, and each of the statistical parameters is normalized to obtain a weight score corresponding to each of the interest tags.

[0105] Specifically, in the embodiment of the present invention, each interest tag can only be configured on one historical user behavior data, and each historical user behavior data can be equipped with multiple interest tags. In addition, the historical user behavior data set may have multiple information of the same type, so each type of interest tag may have multiple. The embodiment of the present invention counts each interest tag through the above judgment process, and finally obtains the weight score corresponding to each interest tag.

[0106] Furthermore, in an embodiment of the present invention, the performing of a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification label corresponding to each historical user behavior data in the historical user behavior data set includes:

[0107] Performing behavior recognition on the user browsing behavior feature sequence to obtain a user behavior recognition result set, and performing a normalization operation on the user behavior recognition result set to obtain a normalized behavior recognition result set;

[0108] According to a pre-built K-means clustering algorithm, a grouping clustering operation based on four clusters is performed on the normalized behavior recognition result set to obtain a behavior summary cluster set;

[0109] Calculating the cluster center position of each behavior summary cluster in the behavior summary cluster set, and arranging each behavior summary cluster according to each cluster center position to obtain a position arrangement result;

[0110] According to the position arrangement result, a behavior classification label is sequentially configured for each of the behavior summary clusters, wherein the behavior classification label includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type.

[0111] Specifically, in an embodiment of the present invention, the user behavior recognition result set in the historical user behavior data set can be first identified, and then the present invention defines a set of behavior classification labels including false touch type, general browsing type, ordinary browsing type and careful browsing type, wherein each label classification is related to the reading time, number of readings, reading speed, etc. in the user behavior recognition result set.

[0112] Then, in the embodiment of the present invention, each user behavior recognition result is normalized by an activation function, and the normalized result is mapped into a space, so as to facilitate the grouping of the user behavior recognition results.

[0113] It should be known that the time change of the user's reading quality from accidental touch to careful reading is not linear, and cannot be simply configured as: 0~0.2 is an accidental touch, 0.2~0.4 is a rough browsing type. Therefore, in the embodiment of the present invention, through the pre-constructed K-means clustering algorithm, by setting the number of clusters to 4, the normalized behavior recognition result set is adaptively grouped into 4 behavior summary clusters, and then the behavior summary clusters are arranged according to the cluster center position of each cluster center point, and according to the front and back relationship in the mapping space, the behavior classification labels of the accidental touch type, rough browsing type, ordinary browsing type and careful browsing type are respectively configured with a behavior summary cluster.

[0114] Thus, the corresponding behavior classification labels are found for each of the historical user behavior data.

[0115] S3. Build a user hobby portrait based on the interest tags and weight scores, obtain a display information set pushed by a pre-built data push service based on the user hobby portrait, and build a browsing behavior habit portrait based on the behavior classification tags corresponding to each of the historical user behavior data.

[0116] In the embodiment of the present invention, the methods of constructing a user hobby portrait and a browsing behavior habit portrait are similar, that is, first collecting tags, then classifying the tags; ensuring the quality of the tags through data cleaning; and then mapping each tag code to the tag set to form a portrait. In addition, other portrait construction technologies, such as model portrait construction, can also be used in the embodiment of the present invention.

[0117] Furthermore, in the embodiment of the present invention, the data push service refers to a service that extracts user-favorite information from massive data and sends it to the user end according to the user's interests and hobbies, which can be used to improve user satisfaction with information.

[0118] S4. Acquire real-time behavior data of the user's response to the display information set, and identify behavior classification labels of the real-time behavior data based on the browsing behavior habit portrait.

[0119] In the embodiment of the present invention, the real-time behavior data of the user terminal is obtained by presetting the embedding point. For example, when a sub-page in a web page is opened, when the URL of the web page changes, it can be recorded to obtain the real-time behavior data. In addition, since the acquisition of the real-time behavior data may be restricted by the information transmission rules, such as starting to collect when the URL changes, the user may have performed multiple user behaviors in the URL, or because the browsing time and other data can only be obtained at the end of the behavior, therefore, one real-time behavior data may contain multiple behaviors.

[0120] In detail, in the embodiment of the present invention, the step of identifying the behavior classification label of the real-time behavior data according to the browsing behavior habit portrait includes:

[0121] Performing transfer learning on the pre-built user behavior classification model according to some network parameters in the user browsing behavior-interest recognition model to obtain an updated user behavior classification model;

[0122] Fine-tune the updated user behavior classification model according to the browsing behavior habit portrait to obtain a local user behavior classification model;

[0123] Using the local user behavior classification model, extracting features from the real-time behavior data to obtain a real-time behavior feature sequence;

[0124] Performing spatial mapping on the real-time behavior feature sequence to obtain the real-time behavior position;

[0125] According to the center positions of each cluster, the real-time behavior position is identified as belonging to a cluster, the belonging behavior summary cluster of the real-time behavior data is obtained, and the behavior classification label corresponding to the belonging behavior summary cluster is obtained.

[0126] Specifically, in the embodiment of the present invention, the user browsing behavior-interest identification model includes the function of identifying user behavior, and after monthly data updates, it is more in line with the user's own situation. Therefore, the new user behavior classification model can be quickly trained directly through transfer learning, and then fine-tuned and trained based on the data combination in the browsing behavior habit portrait to obtain a more accurate local user behavior classification model.

[0127] In the embodiment of the present invention, the feature recognition of the real-time behavior data is realized through the trained local user behavior classification model, so as to obtain the behavior classification label to which the real-time behavior data belongs.

[0128] S5. Determine whether the behavior classification label is the false touch type.

[0129] In the embodiment of the present invention, users are classified by behavioral classification labels, one part of which is the type of accidental touch. This type of data is not beneficial to the optimization of the data push service, so it is not necessary to collect it; while for the other part, it is necessary to configure weights based on factors such as the degree of careful reading by the user.

[0130] When the behavior classification label of the real-time behavior data is the false touch type, S6, determining that the real-time behavior data is invalid;

[0131] When the behavior classification label of the real-time behavior data is a behavior classification label other than the false touch type, S7, weighting the real-time behavior data according to a pre-constructed weight score configuration rule to obtain weighted behavior data.

[0132] In the embodiment of the present invention, according to the pre-constructed weight score configuration rules, for example, the general browsing type, the ordinary browsing type and the detailed browsing type are configured as 0.9, 1 and 1.1 respectively, the real-time behavior data is weighted to obtain weighted behavior data.

[0133] S8. According to the pre-constructed Dirichlet distribution algorithm, a behavior-interest distribution model and an interest-vocabulary distribution model are constructed for the historical user behavior data set, the joint distribution of the behavior-interest distribution model and the interest-vocabulary distribution model is calculated to obtain the behavior-vocabulary distribution model, and the behavior-vocabulary distribution model is used to identify valid information vocabulary in the weighted behavior data.

[0134] In detail, in the embodiment of the present invention, the behavior-interest distribution model and the interest-vocabulary distribution model include:

[0135] The behavior-interest distribution model is expressed as:

[0136]

[0137]

[0138] In the formula, Represents historical user behavior data Distribution of interests, Indicates Historical user behavior data, represents the Dirichlet distribution, represents the total number of historical user behavior data in the historical user behavior data set, represents the prior parameters of the Dirichlet distribution; Indicates the generation of interest tags in the entire historical user behavior data set The probability of Indicates interest tags, represents the Laplace operator, Indicates user Generate interest tags based on historical user behavior data The probability of Represents historical user behavior data;

[0139] The interest-vocabulary distribution model is expressed as:

[0140]

[0141]

[0142] In the formula, Indicate interest tag The vocabulary distribution of Indicates interest tags, represents the total number of interest tags in the historical user behavior data set, represents the prior parameters of the Dirichlet distribution, Represents vocabulary, Represents the vocabulary in the historical user behavior data set The probability of generation.

[0143] Among them, the Dirichlet distribution algorithm is a multivariate probability distribution, which is widely used in fields such as statistics, machine learning and natural language processing, especially in topic modeling.

[0144] The embodiment of the present invention constructs two probability distributions through the Dirichlet distribution algorithm, namely, interest distribution under behavior determination and vocabulary distribution under interest determination.

[0145] Furthermore, in an embodiment of the present invention, the joint distribution of the behavior-interest distribution model and the interest-vocabulary distribution model is calculated to obtain the behavior-vocabulary distribution model.

[0146] In detail, in the embodiment of the present invention, the behavior-vocabulary distribution model is expressed as:

[0147]

[0148] In the formula, Represents a given prior probability and Time, vocabulary and interests distribution.

[0149] In the embodiment of the present invention, the The tag indicating interest is Time Vocabulary The probability of generation, therefore, in the behavior-vocabulary distribution model, , which is equivalent to adding interest tags , assigned to the interest label in the interest distribution .

[0150] Furthermore, in an embodiment of the present invention, valid information words in the weighted behavior data are screened according to a behavior-word distribution model.

[0151] In detail, in the embodiment of the present invention, the use of the behavior-vocabulary distribution model to identify valid information vocabulary in the weighted behavior data includes:

[0152] The behavior-vocabulary distribution model is used to calculate the interest latent content of the weighted behavior data, and the interest latent content is expressed as:

[0153]

[0154] In the formula, Indicates the weighted behavior data The first The implicit interest content corresponding to each word is Representing interest distribution The multinomial distribution of represents the total number of user behaviors in the weighted behavior data, represents the total number of words in the weighted behavioral data;

[0155] The acquisition amount of each word of the weighted behavior data is calculated according to the interest implicit content, and the acquisition amount is expressed as:

[0156]

[0157] In the formula, Indicates the weighted behavior data The first The corresponding acquisition amount of words, Implicit content of interest Multinomial distribution of ;

[0158] According to each of the acquisition amounts, words whose acquisition amounts are greater than a preset qualified threshold are screened to obtain valid information words in the weighted behavior data.

[0159] The acquisition amount is information such as the frequency of occurrence of a word, which can reflect the user's attention to the word and can be changed through configuration. The present invention defines the acquisition amount by indicators such as the number of clicks and browsing time of the user.

[0160] Then, in the embodiment of the present invention, a minimum frequency threshold is set as a qualified threshold, and only words with acquisition amounts higher than the qualified threshold are retained. In this way, those low-frequency words that are hardly paid attention to can be excluded, and valid information words in the weighted behavior data can be obtained.

[0161] S9. Send the valid information vocabulary to the data push service, optimize the user preference portrait, and obtain an optimized information push result.

[0162] In the embodiment of the present invention, the effective information words can be classified and encoded and mapped to the tag set of the previous user preference profile, so as to optimize the user preference profile and obtain an optimized information push result.

[0163] The present invention is to solve the problem described in the background technology. The present invention first extracts and classifies the historical user behavior data set based on user behavior and behavior objects through feature extraction. The user behavior refers to the actions of the user operation, such as reading time, reading speed, reading test, etc., and the behavior object is information such as video and news. These behavior objects will contain corresponding content tags, such as wheat, corn, breeding and pesticides. The present invention can distinguish whether the user is satisfied with the information by analyzing the user behavior, and the analysis of the behavior object can distinguish the user's preference type. The present invention uses a double portrait to set a portrait on the user end and the push end. The browsing behavior habit portrait of the user end is used to collect information recognized by the user, and the push end is used to calculate the user's preferences for push. In addition, when collecting user behavior data, the present invention also further improves the effective information in the user collected data by constructing a behavior-vocabulary distribution model, thereby improving the accuracy of the optimization direction of the push scheme. Therefore, the present invention can solve the problem that the user behavior data collected by the user end is too messy.

[0164] like Figure 2 , which is a functional module diagram of a push solution optimization system based on user-side response data collection provided by an embodiment of the present invention.

[0165] The push scheme optimization system 100 based on user-side response data collection of the present invention can be installed in an electronic device. According to the functions implemented, the push scheme optimization system 100 based on user-side response data collection can include a feature extraction module 101, a feature classification module 102, a dual-portrait construction module 103, an effective user data collection module 104 and a push optimization module 105. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0166] The feature extraction module 101 is used to obtain a historical user behavior data set within a preset time period, obtain a historical user behavior data set within a preset time period, and use a pre-trained user browsing behavior-interest recognition model to perform a feature extraction operation based on user behavior and behavior objects on the historical user behavior data set to obtain a user browsing behavior feature sequence and a user interest feature sequence;

[0167] The feature classification module 102 is used to perform a quantitative statistical operation on the user interest feature sequence based on interest tags to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag, and perform a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification label corresponding to each historical user behavior data in the historical user behavior data set, wherein the behavior classification label includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type;

[0168] The dual-profile construction module 103 is used to construct a user hobby portrait according to the interest tags and weight scores, and obtain a display information set pushed by a pre-built data push service according to the user hobby portrait, and construct a browsing behavior habit portrait according to the behavior classification tags corresponding to each of the historical user behavior data;

[0169] The effective user data acquisition module 104 is used to obtain the real-time behavior data of the user's response to the display information set, and identify the behavior classification label of the real-time behavior data according to the browsing behavior habit portrait, and when the behavior classification label of the real-time behavior data is the mis-touch type, determine that the real-time behavior data is invalid, and when the behavior classification label of the real-time behavior data is other behavior classification labels other than the mis-touch type, perform weight configuration on the real-time behavior data according to the pre-constructed weight score configuration rule to obtain weighted behavior data, and construct a behavior-interest distribution model and an interest-lexicon distribution model for the historical user behavior data set according to the pre-constructed Dirichlet distribution algorithm, calculate the joint distribution of the behavior-interest distribution model and the interest-lexicon distribution model to obtain the behavior-lexicon distribution model, and use the behavior-lexicon distribution model to identify effective information vocabulary in the weighted behavior data;

[0170] The push optimization module 105 is used to send the valid information vocabulary to the data push service, optimize the user preference portrait, and obtain an optimized information push result.

[0171] In detail, the modules in the push scheme optimization system 100 based on user terminal response data collection in the embodiment of the present invention are used in the same manner as described above. Figure 1 The push scheme optimization method based on user-side response data collection described in the above is the same technical means and can produce the same technical effects, so it will not be repeated here.

[0172] like Figure 3 , is a schematic diagram of the structure of an electronic device for implementing a push solution optimization method based on user-side response data collection provided by an embodiment of the present invention.

[0173] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a push scheme optimization method program based on user-side response data collection.

[0174] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 also includes an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the push scheme optimization method program based on the user-side response data collection, but also can be used to temporarily store data that has been output or is to be output.

[0175] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as a push scheme optimization method program based on user-side response data collection, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0176] The bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0177] Figure 3 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 3The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0178] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0179] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0180] Optionally, the electronic device 1 may further include a user interface, which may be a display (Display), an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device, etc. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0181] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0182] The push scheme optimization method program based on user terminal response data collection stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0183] Obtain a historical user behavior data set within a preset time period, and use a pre-trained user browsing behavior-interest recognition model to perform a feature extraction operation based on user behavior and behavior objects on the historical user behavior data set to obtain a user browsing behavior feature sequence and a user interest feature sequence;

[0184] Performing a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag, and performing a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification tag corresponding to each historical user behavior data in the historical user behavior data set, wherein the behavior classification tag includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type;

[0185] According to the interest tags and weight scores, a user hobby portrait is constructed, and a display information set pushed by a pre-constructed data push service according to the user hobby portrait is obtained, and a browsing behavior habit portrait is constructed according to the behavior classification tags corresponding to each of the historical user behavior data;

[0186] Acquire real-time behavior data of the user's response to the display information set, and identify behavior classification labels of the real-time behavior data according to the browsing behavior habit portrait;

[0187] When the behavior classification label of the real-time behavior data is the false touch type, determining that the real-time behavior data is invalid;

[0188] When the behavior classification label of the real-time behavior data is a behavior classification label other than the false touch type, weighting the real-time behavior data according to a pre-constructed weight score configuration rule to obtain weighted behavior data;

[0189] According to a pre-constructed Dirichlet distribution algorithm, a behavior-interest distribution model and an interest-lexicon distribution model are constructed for the historical user behavior data set, a joint distribution of the behavior-interest distribution model and the interest-lexicon distribution model is calculated to obtain a behavior-lexicon distribution model, and valid information words in the weighted behavior data are identified using the behavior-lexicon distribution model;

[0190] The valid information vocabulary is sent to the data push service, the user preference portrait is optimized, and an optimized information push result is obtained.

[0191] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0192] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0193] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0194] Obtain a historical user behavior data set within a preset time period, and use a pre-trained user browsing behavior-interest recognition model to perform a feature extraction operation based on user behavior and behavior objects on the historical user behavior data set to obtain a user browsing behavior feature sequence and a user interest feature sequence;

[0195] Performing a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag, and performing a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification tag corresponding to each historical user behavior data in the historical user behavior data set, wherein the behavior classification tag includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type;

[0196] According to the interest tags and weight scores, a user hobby portrait is constructed, and a display information set pushed by a pre-constructed data push service according to the user hobby portrait is obtained, and a browsing behavior habit portrait is constructed according to the behavior classification tags corresponding to each of the historical user behavior data;

[0197] Acquire real-time behavior data of the user's response to the display information set, and identify behavior classification labels of the real-time behavior data according to the browsing behavior habit portrait;

[0198] When the behavior classification label of the real-time behavior data is the false touch type, determining that the real-time behavior data is invalid;

[0199] When the behavior classification label of the real-time behavior data is a behavior classification label other than the false touch type, weighting the real-time behavior data according to a pre-constructed weight score configuration rule to obtain weighted behavior data;

[0200] According to a pre-constructed Dirichlet distribution algorithm, a behavior-interest distribution model and an interest-lexicon distribution model are constructed for the historical user behavior data set, a joint distribution of the behavior-interest distribution model and the interest-lexicon distribution model is calculated to obtain a behavior-lexicon distribution model, and valid information words in the weighted behavior data are identified using the behavior-lexicon distribution model;

[0201] The valid information vocabulary is sent to the data push service, the user preference portrait is optimized, and an optimized information push result is obtained.

[0202] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, and actual implementation may have other division methods.

[0203] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0204] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0205] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0206] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.

[0207] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for optimizing a push scheme based on user-side response data collection, characterized in that: The method comprises: Obtain a set of historical user behavior data within a preset time period; Using a pre-trained user browsing behavior-interest recognition model, the historical user behavior data set is quantized and encoded to obtain a historical behavior feature vector set, and each historical behavior feature vector in the historical behavior feature vector set is concatenated with a vector of a preset character length to obtain a historical behavior feature matrix; Using the pre-constructed convolution kernel set in the user browsing behavior-interest recognition model, a convolution operation is performed on the historical behavior feature matrix to obtain a convolution matrix set; Using the pre-constructed pooling network in the user browsing behavior-interest recognition model, a maximum pooling operation is performed on the convolution matrix set to obtain a pooling matrix set; Using the pre-constructed flattened network in the user browsing behavior-interest recognition model, the pooling matrix set is subjected to feature concatenation and dimensionality reduction operations to obtain a historical behavior feature sequence; Using the fully connected network pre-constructed in the user browsing behavior-interest recognition model, the historical behavior feature sequence is subjected to a binary classification operation based on user behavior and behavior object, thereby obtaining a user browsing behavior feature sequence and a user interest feature sequence respectively; Performing a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag, and performing a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification tag corresponding to each historical user behavior data in the historical user behavior data set, wherein the behavior classification tag includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type; According to the interest tags and weight scores, a user hobby portrait is constructed, and a display information set pushed by a pre-constructed data push service according to the user hobby portrait is obtained, and a browsing behavior habit portrait is constructed according to the behavior classification tags corresponding to each of the historical user behavior data; Acquire real-time behavior data of the user's response to the display information set, and identify behavior classification labels of the real-time behavior data according to the browsing behavior habit portrait; When the behavior classification label of the real-time behavior data is the false touch type, determining that the real-time behavior data is invalid; When the behavior classification label of the real-time behavior data is a behavior classification label other than the false touch type, weighting the real-time behavior data according to a pre-constructed weight score configuration rule to obtain weighted behavior data; According to the pre-constructed Dirichlet distribution algorithm, a behavior-interest distribution model and an interest-lexicon distribution model are constructed for the historical user behavior data set, and the joint distribution of the behavior-interest distribution model and the interest-lexicon distribution model is calculated to obtain a behavior-lexicon distribution model. The behavior-lexicon distribution model is used to identify valid information words in the weighted behavior data, wherein the behavior-interest distribution model is expressed as: In the formula, Represents historical user behavior data Distribution of interests, Indicates Historical user behavior data, represents the Dirichlet distribution, represents the total number of historical user behavior data in the historical user behavior data set, represents the prior parameters of the Dirichlet distribution, Indicates the generation of interest tags in the entire historical user behavior data set The probability of Indicates interest tags, represents the Laplace operator, Indicates user Generate interest tags based on historical user behavior data The probability of Represents historical user behavior data; The interest-vocabulary distribution model is expressed as: In the formula, Indicate interest tag The vocabulary distribution of Indicates interest tags, represents the total number of interest tags in the historical user behavior data set, represents the prior parameters of the Dirichlet distribution, Represents vocabulary, Represents the vocabulary in the historical user behavior data set Probability of generation; The behavior-vocabulary distribution model is expressed as: In the formula, It means that the given prior probability is and Time, vocabulary and interests Distribution of The valid information vocabulary is sent to the data push service, the user preference portrait is optimized, and an optimized information push result is obtained.

2. The push scheme optimization method based on user terminal response data collection according to claim 1 is characterized in that: The step of obtaining a set of historical user behavior data within a preset time period includes: Integrate historical user behavior data within a preset time period from multiple data sources to obtain an initial data set; Performing a duplicate check on the initial data set, deleting duplicate behavior data in the initial data set to obtain a clean data set, and performing data standardization on the clean data set to obtain a standard data set, wherein the data standardization includes lowercase conversion, stop word removal, and Z-score standardization; The TF-IDF index of each word in the standard data set is calculated, and important features of each word in the standard data set are extracted according to the TF-IDF index to obtain a historical user behavior data set.

3. The push scheme optimization method based on user terminal response data collection according to claim 2 is characterized in that: The performing a quantitative statistical operation based on interest tags on the user interest feature sequence to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag includes: Performing feature recognition on the user interest feature sequence to obtain an interest tag set; Marking each interest tag in the interest tag set on the corresponding historical user behavior data, and equipping each interest tag with statistical parameters; Determine whether each of the interest tags is marked for the first time in the corresponding historical user behavior data; When the interest tag is marked for the first time, adding one to the statistical parameter of the interest tag; When the interest tag is not marked for the first time, a zero addition operation is performed on the statistical parameter of the interest tag; Statistical parameters corresponding to each interest tag are obtained, and each of the statistical parameters is normalized to obtain a weight score corresponding to each of the interest tags.

4. The push scheme optimization method based on user terminal response data collection according to claim 3 is characterized in that: The performing a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification label corresponding to each historical user behavior data in the historical user behavior data set includes: Performing behavior recognition on the user browsing behavior feature sequence to obtain a user behavior recognition result set, and performing a normalization operation on the user behavior recognition result set to obtain a normalized behavior recognition result set; According to a pre-built K-means clustering algorithm, a grouping clustering operation based on four clusters is performed on the normalized behavior recognition result set to obtain a behavior summary cluster set; Calculating the cluster center position of each behavior summary cluster in the behavior summary cluster set, and arranging each behavior summary cluster according to each cluster center position to obtain a position arrangement result; According to the position arrangement result, a behavior classification label is sequentially configured for each of the behavior summary clusters, wherein the behavior classification label includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type.

5. The push scheme optimization method based on user terminal response data collection according to claim 4 is characterized in that: The step of identifying the behavior classification label of the real-time behavior data according to the browsing behavior habit portrait includes: Performing transfer learning on the pre-built user behavior classification model according to some network parameters in the user browsing behavior-interest recognition model to obtain an updated user behavior classification model; Fine-tune the updated user behavior classification model according to the browsing behavior habit portrait to obtain a local user behavior classification model; Using the local user behavior classification model, extracting features from the real-time behavior data to obtain a real-time behavior feature sequence; Performing spatial mapping on the real-time behavior feature sequence to obtain the real-time behavior position; According to the center positions of each cluster, the real-time behavior position is identified as belonging to a cluster, the belonging behavior summary cluster of the real-time behavior data is obtained, and the behavior classification label corresponding to the belonging behavior summary cluster is obtained.

6. The push scheme optimization method based on user terminal response data collection according to claim 5 is characterized in that: The step of using the behavior-vocabulary distribution model to identify valid information vocabulary in the weighted behavior data includes: The behavior-vocabulary distribution model is used to calculate the interest latent content of the weighted behavior data, and the interest latent content is expressed as: In the formula, Indicates the weighted behavior data The first The implicit interest content corresponding to each word is Representing interest distribution The multinomial distribution of represents the total number of user behaviors in the weighted behavior data, represents the total number of words in the weighted behavioral data; The acquisition amount of each word of the weighted behavior data is calculated according to the interest implicit content, and the acquisition amount is expressed as: In the formula, Indicates the weighted behavior data The first The corresponding acquisition amount of words, Implicit content of interest Multinomial distribution of ; According to each of the acquisition amounts, words whose acquisition amounts are greater than a preset qualified threshold are screened to obtain valid information words in the weighted behavior data.

7. A push scheme optimization system based on user terminal response data collection, the system is suitable for implementing the push scheme optimization method based on user terminal response data collection according to claim 1, the system comprising: A feature extraction module is used to obtain a historical user behavior data set within a preset time period, obtain a historical user behavior data set within a preset time period, and use a pre-trained user browsing behavior-interest recognition model to perform a feature extraction operation based on user behavior and behavior objects on the historical user behavior data set to obtain a user browsing behavior feature sequence and a user interest feature sequence; A feature classification module is used to perform a quantitative statistical operation on the user interest feature sequence based on interest tags to obtain each interest tag that the user is interested in and a weight score corresponding to each interest tag, and to perform a behavior classification operation on the user browsing behavior feature sequence to obtain a behavior classification label corresponding to each historical user behavior data in the historical user behavior data set, wherein the behavior classification label includes an accidental touch type, a general browsing type, a normal browsing type, and a careful browsing type; A dual-profile construction module is used to construct a user hobby portrait based on the interest tags and weight scores, obtain a display information set pushed by a pre-built data push service based on the user hobby portrait, and construct a browsing behavior habit portrait based on the behavior classification tags corresponding to each of the historical user behavior data; An effective user data collection module is used to obtain the real-time behavior data of the user's response to the display information set, and identify the behavior classification label of the real-time behavior data according to the browsing behavior habit portrait, and when the behavior classification label of the real-time behavior data is the mis-touch type, determine that the real-time behavior data is invalid, and when the behavior classification label of the real-time behavior data is other behavior classification labels other than the mis-touch type, perform weight configuration on the real-time behavior data according to a pre-constructed weight score configuration rule to obtain weighted behavior data, and construct a behavior-interest distribution model and an interest-lexicon distribution model for the historical user behavior data set according to a pre-constructed Dirichlet distribution algorithm, calculate the joint distribution of the behavior-interest distribution model and the interest-lexicon distribution model to obtain a behavior-lexicon distribution model, and use the behavior-lexicon distribution model to identify effective information vocabulary in the weighted behavior data; The push optimization module is used to send the valid information vocabulary to the data push service, optimize the user preference portrait, and obtain an optimized information push result.