Intelligent applet recommendation method and device based on function summary

Abstract

The embodiment of the invention provides an applet intelligent recommendation method and device based on a function summary, and the method comprises the steps: carrying out the cutting of text description and code data of an applet through employing a preset text analysis and mode recognition algorithm, and obtaining the cutting data, extracting keywords and functional features from the block data through a preset natural language processing algorithm, and determining corresponding applet function summary features according to the keywords and the functional features; determining a corresponding applet pushing model according to the applet classification features and a back propagation algorithm, and executing applet pushing operation according to the applet pushing model; according to the method and the device, the applet pushing accuracy can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for intelligently recommending mini-programs based on function summaries. Background Art

[0002] In the current field of digital marketing and user experience optimization, mini programs have become an important application tool for major platforms and enterprises due to their lightweight, convenient and rich functions. With the help of mini programs, enterprises can reach users more efficiently, provide personalized services and improve user stickiness. However, with the rapid increase in the number of mini programs and the diversification of user needs, how to achieve accurate recommendations and intelligent push has become a key issue in improving user experience and marketing effectiveness. Although the application of existing technologies in this field has made some progress, it still faces many challenges and shortcomings.

[0003] Existing mini-program recommendation systems usually rely on simple rule matching and user behavior analysis. By analyzing the user's historical access records, click behavior, usage time and other data, the system can preliminarily judge the user's interests and needs and make recommendations accordingly. However, this method is limited in capturing user needs and is difficult to fully and accurately reflect the user's real interests. Especially when facing new users or insufficient historical data, the recommendation effect is often unsatisfactory. In addition, simple behavior analysis methods are difficult to cope with the dynamic changes in user needs, and the timeliness and relevance of recommendation results are low.

[0004] On the other hand, existing mini-program recommendation systems often use manual labeling or simple keyword matching methods when processing the functional features of mini-programs. This method is not only inefficient, but also prone to inaccurate labeling or unreasonable keyword selection. As mini-program functions become increasingly complex and diverse, manual labeling methods are obviously difficult to meet the needs. Even methods based on keyword matching are difficult to fully capture the functional features of mini-programs, resulting in limited accuracy and coverage of recommendation results.

[0005] In terms of model selection and optimization, existing mini-program recommendation systems usually use traditional machine learning models, such as decision trees and random forests. Although these models can improve the accuracy of recommendations to a certain extent, they still have performance bottlenecks when processing high-dimensional data and complex features. In particular, decision tree models are prone to overfitting or underfitting problems during feature selection and splitting, which affects the generalization ability and stability of the model. In addition, the existing models have low computational efficiency when processing large-scale data, making it difficult to respond to user needs in real time.

[0006] In addition, the existing mini-program recommendation system lacks effective model integration and optimization mechanisms. Although some systems have tried to introduce integrated learning methods, such as gradient boosting and random forest, the integration effect and computational efficiency of the model still need to be improved in practical applications. Especially in the case of multi-source heterogeneous data, how to effectively integrate the prediction results of different models and improve the accuracy and robustness of recommendations is an urgent problem to be solved.

[0007] Existing technologies also have deficiencies in feedback and optimization of recommendation results. Most existing systems rely on static rules and preset logic, lacking the ability to dynamically adjust and self-optimize. Due to the constant changes in user needs and market environment, fixed recommendation strategies are difficult to maintain long-term effectiveness. In addition, during the push process, existing systems often ignore user feedback on recommended content, lack effective feedback mechanisms and model adjustment methods, resulting in difficulty in continuously optimizing recommendation results and affecting user experience.

[0008] In summary, the existing mini-program recommendation system still has many deficiencies and challenges in terms of capturing user needs, extracting functional features, selecting and optimizing models, and feedback and adjustment of recommendation results. Faced with increasingly complex user needs and market environment, a more intelligent, accurate, and efficient mini-program recommendation method is urgently needed to improve user experience and marketing effectiveness. Summary of the invention

[0009] In response to the problems in the prior art, the present application provides a method and device for intelligently recommending mini-programs based on functional summaries, which can effectively improve the accuracy of mini-program push.

[0010] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0011] In a first aspect, the present application provides a method for intelligently recommending mini-programs based on a function summary, comprising:

[0012] Using a preset text analysis and pattern recognition algorithm to slice the text description and code data of the mini program to obtain sliced ​​data, extracting keywords and functional features from the sliced ​​data using a preset natural language processing algorithm, and determining corresponding mini program functional summary features based on the keywords and functional features;

[0013] Input the mini-program function summary features into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic features, input the semantic features into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency features, input the semantic dependency features into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification features;

[0014] The corresponding mini-program push model is determined according to the mini-program classification features and the back-propagation algorithm, and the mini-program push operation is performed according to the mini-program push model.

[0015] Furthermore, the determining of the corresponding mini-program function summary features according to the keywords and function features includes:

[0016] Performing a minority class recognition operation on the keywords and functional features according to a preset minority class recognition rule, and determining the corresponding nearest neighbor mini program functional features according to the minority class mini program functional features obtained after the minority class recognition operation and a preset K nearest neighbor algorithm;

[0017] The corresponding mini-program function summary features are determined according to the minority mini-program function features and the nearest neighbor mini-program function features.

[0018] Furthermore, the determining of the corresponding nearest neighbor mini-program functional features based on the minority mini-program functional features obtained after the minority class identification operation and a preset K nearest neighbor algorithm includes:

[0019] Performing a distance calculation operation on the minority mini-program functional features and the nearest neighbor mini-program functional features to determine the corresponding mini-program functional feature distances;

[0020] The corresponding synthesized mini-program functional features are determined according to the mini-program functional feature distance and a preset ratio rule, and the mini-program functional features are subjected to feature addition operations according to the synthesized mini-program functional features to determine the corresponding mini-program functional summary features.

[0021] Furthermore, the step of inputting the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation to determine the corresponding semantic features includes:

[0022] Input the conversion code into the bidirectional encoder layer of the setting information extraction model, perform a vector conversion operation on the conversion code according to a preset word embedding algorithm, and determine a corresponding encoding vector;

[0023] A coding stacking operation is performed according to the coding vector and a preset number of coding structure layers to determine corresponding semantic features.

[0024] Furthermore, the step of inputting the semantic features into the bidirectional long short-term memory network layer of the setting information extraction model to perform a dependency capture operation and determine the corresponding semantic dependency features includes:

[0025] Inputting the semantic features into the bidirectional long short-term memory network layer of the setting information extraction model, performing dependency capture operations according to a preset gating mechanism, and determining a corresponding semantic feature matrix;

[0026] The corresponding semantic dependency features are determined according to the semantic feature matrix and the preset memory unit.

[0027] Furthermore, the step of inputting the semantic dependency feature into the conditional random field layer of the setting information extraction model to determine the corresponding mini-program classification feature includes:

[0028] Inputting the semantic dependency feature into the conditional random field layer of the setting information extraction model to determine a label sequence corresponding to the semantic dependency feature;

[0029] The corresponding mini-program classification features are determined according to the tag sequence and a preset Viterbi algorithm.

[0030] Furthermore, the determining of the corresponding mini program classification features further includes:

[0031] Perform a weight matching operation based on the decision tree prediction features and the regression tree prediction features output by the pruning layer in the preset regression tree model to determine the corresponding decision tree prediction feature weights and regression tree prediction feature weights;

[0032] A feature weighted averaging operation is performed based on the decision tree prediction feature weights and the regression tree prediction feature weights to determine the corresponding mini-program classification features.

[0033] In a second aspect, the present application provides a device for intelligently recommending mini-programs based on a functional summary, comprising:

[0034] A user portrait collection module is used to collect historical mini-program access data sent by users, slice the text description and code data of the mini-program using a preset text analysis and pattern recognition algorithm to obtain sliced ​​data, extract keywords and functional features from the sliced ​​data using a preset natural language processing algorithm, and determine the corresponding mini-program functional summary features based on the keywords and functional features;

[0035] A mini-program classification feature module is used to input the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic feature, input the semantic feature into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency feature, input the semantic dependency feature into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification feature;

[0036] The mini-program push module is used to determine the corresponding mini-program push model according to the mini-program classification characteristics and the back propagation algorithm, and perform the mini-program push operation according to the mini-program push model.

[0037] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for intelligently recommending mini-programs based on functional summaries when executing the program.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for intelligently recommending mini-programs based on functional summaries.

[0039] In a fifth aspect, the present application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the method for intelligent recommendation of mini-programs based on functional summaries.

[0040] It can be seen from the above technical scheme that the present application provides a method and device for intelligent recommendation of mini-programs based on functional summaries, which obtains segmented data by using preset text analysis and pattern recognition algorithms to segment the text description and code data of the mini-programs, and extracts keywords and functional features from the segmented data by using a preset natural language processing algorithm, and determines the corresponding mini-program functional summary features based on the keywords and functional features; determines the corresponding mini-program push model based on the mini-program classification features and the back propagation algorithm, and executes the mini-program push operation based on the mini-program push model; the present application can effectively improve the accuracy of mini-program push. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 This is one of the flowcharts of the method for intelligently recommending mini-programs based on function summaries in an embodiment of the present application;

[0043] Figure 2 This is the second flow chart of the method for intelligently recommending mini-programs based on function summaries in an embodiment of the present application;

[0044] Figure 3 This is a flowchart of the third method for intelligently recommending mini-programs based on function summaries in an embodiment of the present application;

[0045] Figure 4 This is a fourth flowchart of the method for intelligently recommending mini-programs based on function summaries in an embodiment of the present application;

[0046] Figure 5 This is a fifth flow chart of the method for intelligently recommending mini-programs based on function summaries in an embodiment of the present application;

[0047] Figure 6 This is a sixth flow chart of the method for intelligently recommending mini-programs based on function summaries in an embodiment of the present application;

[0048] Figure 7 This is the seventh flow chart of the method for intelligently recommending mini-programs based on function summaries in the embodiment of the present application;

[0049] Figure 8 This is a structural diagram of a device for intelligently recommending small programs based on a functional summary in an embodiment of the present application;

[0050] Fig. 9 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application.

[0051] Reference numerals:

[0052] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver program storage unit 9144, antenna 9111, speaker 9131, microphone 9132. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0054] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0055] In view of the problems existing in the prior art, the present application provides a method and device for intelligently recommending mini-programs based on functional summaries, which obtains segmented data by using preset text analysis and pattern recognition algorithms, extracts keywords and functional features from the segmented data by using a preset natural language processing algorithm, and determines the corresponding mini-program functional summary features based on the keywords and functional features; determines the corresponding mini-program push model based on the mini-program classification features and the back propagation algorithm, and executes the mini-program push operation based on the mini-program push model; the present application can effectively improve the accuracy of mini-program push.

[0056] In order to effectively improve the accuracy of mini program push, the present application provides an embodiment of a mini program intelligent recommendation method based on function summary, see Figure 1 The mini-program intelligent recommendation method based on function summary specifically includes the following contents:

[0057] Step S101: using a preset text analysis and pattern recognition algorithm to segment the text description and code data of the mini program to obtain segmented data, extracting keywords and functional features from the segmented data using a preset natural language processing algorithm, and determining corresponding mini program functional summary features based on the keywords and functional features;

[0058] Optionally, in this embodiment, in step S101, the system needs to use a preset text analysis and pattern recognition algorithm to slice the text description and code data of the mini program to obtain sliced ​​data. Then, the preset natural language processing algorithm is used to extract keywords and functional features from these sliced ​​data, and finally the corresponding mini program functional summary features are determined based on these keywords and functional features. The specific technical process is as follows:

[0059] First, the system will preprocess the text description and code data of the mini program. The text description usually includes human-readable text information such as the function introduction and instructions for use of the mini program; the code data includes the actual implementation code of the mini program. The purpose of preprocessing is to convert these different forms of data into a unified format that is convenient for subsequent processing. After the preprocessing is completed, the system will use the preset text analysis and pattern recognition algorithms to cut these data into pieces. Chunking is a technology that divides text and code into small segments so that each small segment (chunk) can be analyzed independently. The granularity of the chunks can be adjusted according to actual needs, and can be divided by line or by functional module.

[0060] After the segmented data is generated, the system will apply the preset natural language processing (NLP) algorithm to extract keywords and functional features from these segmented data. NLP algorithms include technologies such as word segmentation, part-of-speech tagging, and named entity recognition. Through these technologies, the system can identify key nouns, verbs, and other information from text descriptions, and identify important programming-related features such as function names and variable names from code data. The purpose of keyword extraction is to find out those words that can represent the main functions and characteristics of the mini program, while functional feature extraction is to identify the functional modules and operation logic specifically implemented by the mini program.

[0061] In terms of technical principles, text analysis and pattern recognition algorithms utilize the basic principles of natural language processing and code parsing to extract meaningful feature information through structured analysis of text and code. The NLP algorithm further processes this feature information to extract keywords and functional features. These keywords and functional features can accurately reflect the core functions and characteristics of the mini program, providing a basis for the subsequent determination of functional profile features.

[0062] In terms of the technical problems solved, step S101 solves the problem of how to efficiently and accurately extract functional features from the text description and code data of the mini-program. In practical applications, text descriptions and code data are usually highly heterogeneous and complex, and it is very challenging to directly extract feature information from them. Through block processing and NLP algorithms, the system can effectively reduce the complexity of the data and improve the accuracy and efficiency of feature extraction.

[0063] In terms of the technical effects achieved, through step S101, the system can extract accurate keywords and functional features from the text description and code data of the mini-program, and determine the functional summary features of the mini-program based on this information. This not only improves the accuracy of feature extraction, but also enhances the degree of automation of feature extraction and reduces the need for manual intervention.

[0064] For example, on a mini-program development platform, a developer uploaded a to-do list mini-program for managing personal tasks. The text description may include content such as "the mini-program can add, delete and edit to-do items, support setting reminders and marking completion status", and the code data includes specific codes to implement these functions. Through step S101, the system can cut these text descriptions and code data into pieces, extract keywords and functional features such as "add to-do items", "delete to-do items", "set reminders", and finally determine that the functional summary feature of the mini-program is "task management".

[0065] In another example, suppose a mini program is used for online education, providing course playback and online test functions. The text description may include content such as "users can watch video courses, complete post-course tests, and view test scores", and the code data includes specific codes to implement these functions. Through step S101, the system can extract keywords and functional features such as "video courses", "online tests", and "view scores", and determine that the functional summary feature of the mini program is "online education".

[0066] In general, through step S101, the system can efficiently and accurately extract functional features from the text description and code data of the mini-program, and determine the functional summary features of the mini-program based on these features. This process not only improves the automation level of feature extraction, but also improves the accuracy and efficiency of feature extraction, and is widely used in various mini-program development and management platforms that require automated analysis and feature extraction. Through this method, developers can more easily understand the functional characteristics of mini-programs, and users can also find mini-programs that meet their needs more quickly.

[0067] Step S102: Input the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic feature, input the semantic feature into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency feature, input the semantic dependency feature into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification feature;

[0068] Optionally, in this embodiment, in step S102, the system inputs the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation to determine the corresponding conversion code. The mini-program function summary feature usually includes text information that briefly describes the mini-program function. This information needs to be pre-processed and encoded so that the subsequent model can efficiently process and understand it.

[0069] In the specific implementation process, the system first pre-processes the features of the Mini Program function summary, including steps such as text cleaning, word segmentation and vectorization. Text cleaning refers to the removal of irrelevant information and noise data, such as punctuation marks, stop words, etc. Word segmentation refers to the division of text into individual words or phrases for subsequent vectorization processing. The vectorization method can use word embedding technology, such as Word2Vec, GloVe or BERT, to convert text information into fixed-length vector representations, which are conversion codes.

[0070] Next, the system inputs the converted code into the bidirectional encoder layer of the information extraction model for semantic transformation. The bidirectional encoder layer usually refers to a bidirectional recurrent neural network (Bi-RNN) or a bidirectional long short-term memory network (Bi-LSTM), which can capture the contextual semantic relationship of text information through forward and backward propagation. Through the processing of the bidirectional encoder layer, the system can generate richer semantic feature representations.

[0071] The technical principle is mainly based on natural language processing (NLP) and deep learning technology. Word embedding technology can convert text information into high-dimensional vector representation and capture the semantic information of words. Bidirectional recurrent neural network or bidirectional long short-term memory network captures the contextual semantic relationship of text through forward and backward propagation, improving the accuracy and completeness of semantic feature representation.

[0072] The generated semantic features need to be further processed to capture the dependencies in the text information. The system inputs the semantic features into the bidirectional long short-term memory network layer (Bi-LSTM Layer) of the set information extraction model to perform dependency capture operations. The bidirectional long short-term memory network can effectively capture the long and short dependencies in the text sequence through memory units and gating mechanisms, thereby generating more accurate semantic dependency features.

[0073] The technical principle of dependency capture operation is mainly based on the long short-term memory network (LSTM) and its bidirectional extension (Bi-LSTM). LSTM can solve the gradient vanishing and gradient exploding problems of traditional recurrent neural network (RNN) when processing long sequences through memory units and gating mechanisms. Bidirectional LSTM further enhances the ability to capture dependencies in text sequences through forward and backward propagation.

[0074] Finally, the system inputs the semantic dependency features into the conditional random field layer (CRF Layer) of the setting information extraction model to determine the corresponding mini-program classification features. Conditional random field is a probabilistic graph model suitable for sequence labeling tasks, which can improve the accuracy of classification by considering contextual information. The CRF layer calculates the conditional probability of the semantic dependency features and determines the classification labels at each position in the text, thereby generating mini-program classification features.

[0075] The technical principle is mainly based on the conditional random field (CRF) model. CRF can improve the accuracy and robustness of sequence annotation by modeling the global dependencies in feature sequences. In this solution, the CRF layer determines the classification labels of the mini-program function summary features by calculating the conditional probability of the semantic dependency features.

[0076] This solution effectively solves the problems of encoding conversion, semantic transformation, dependency capture and classification feature determination of mini-program function summary features. Through the combination of multi-layer neural network and conditional random field model, the system can efficiently and accurately process and classify mini-program function summary features, providing a reliable data foundation for subsequent mini-program management and recommendation.

[0077] In terms of technical effects, the system uses a multi-layer neural network and conditional random field model to achieve automatic information extraction and classification of the summary features of mini-program functions. Encoding conversion and semantic transformation operations can generate rich semantic feature representations, dependency capture operations can effectively identify long and short dependencies in text, and classification feature determination operations can efficiently and accurately generate mini-program classification features. The combination of these technical means not only improves the accuracy and efficiency of information extraction and classification, but also provides reliable data support for subsequent mini-program management and recommendations.

[0078] For example, suppose the user is an operator of a mini-program development platform and needs to classify and manage the functions of the mini-programs on the platform. The user enters the functional summary features of the mini-program, such as "This is a mini-program that helps users record their daily steps and health data." The system first preprocesses and transcodes these text information to generate the corresponding conversion code. Next, the system inputs the conversion code into the bidirectional encoder layer to generate a semantic feature representation. Then, the system inputs the semantic features into the bidirectional long short-term memory network layer to capture the dependencies in the text and generate semantic dependency features. Finally, the system inputs the semantic dependency features into the conditional random field layer to determine the classification features of the mini-program, such as "health management", "exercise records", etc. Through these steps, the system can efficiently and accurately classify the functions of mini-programs, helping operators to better manage and recommend mini-programs.

[0079] To summarize, step S102 solves the problems of encoding conversion, semantic transformation, dependency capture and classification feature determination of the mini-program function summary features through a multi-layer neural network and conditional random field model, improves the accuracy and efficiency of information extraction and classification, and provides reliable data support for mini-program management and recommendation.

[0080] Step S103: Determine a corresponding mini-program push model according to the mini-program classification features and the back propagation algorithm, and perform a mini-program push operation according to the mini-program push model.

[0081] Optionally, in this embodiment, in step S103, the system determines the corresponding mini program push model based on the mini program classification features and the back propagation algorithm, and performs the mini program push operation based on the model. The mini program classification features are generated by the previous steps, which represent the functions and uses of the mini program and reflect the applicability of the mini program in different user demand scenarios.

[0082] In the specific implementation process, the system first needs to analyze and process the classification features of the Mini Program in order to provide accurate input for the subsequent Mini Program push model. Mini Program classification features usually include multi-dimensional information, such as functional categories, user reviews, usage frequency, etc. The system converts these classification features into a format acceptable to the model through feature engineering and data preprocessing. The feature engineering process may include steps such as feature selection, feature scaling, and feature combination to improve the performance and accuracy of the model.

[0083] Next, the system uses the backpropagation algorithm to train the mini-program push model. The backpropagation algorithm is a major training method in deep learning. It calculates the error between the model output and the true value, and propagates the error back to each layer of the network to update the weight parameters, thereby continuously optimizing the model performance. In this solution, the system uses the mini-program classification features as input and trains a recommendation model through the backpropagation algorithm, so that it can accurately predict users' preferences for different mini-programs.

[0084] The technical principle is mainly based on deep neural networks and back propagation algorithms. Deep neural networks contain multiple hidden layers and can capture complex patterns and features in data through multi-layer nonlinear transformations. The back propagation algorithm calculates and adjusts network weights layer by layer through the gradient descent method to minimize the loss function, thereby improving the prediction accuracy of the model. During the training process, the system will continuously iteratively adjust the model parameters until the loss function converges to obtain a mini program push model with better performance.

[0085] After the mini program push model is determined, the system executes the mini program push operation according to the model. The push operation refers to recommending the most suitable mini program to the user based on the user's preferences and behavioral characteristics. The system first needs to collect the user's behavioral data, such as browsing history, click history, historical usage, etc., and input this data into the mini program push model. The model generates a push list through a comprehensive analysis of the user's behavioral data and the mini program classification characteristics, and recommends the mini programs that the user may be interested in.

[0086] This solution effectively solves the problem of personalization and accuracy of mini-program recommendations. By combining the mini-program classification features and user behavior data, the system can generate a personalized recommendation list to improve the relevance of push notifications and user satisfaction. The use of the back-propagation algorithm enables the model to be continuously optimized and improved, gradually improving the accuracy of recommendations.

[0087] In terms of technical effects, the system uses the back-propagation algorithm to train the mini-program push model and achieve personalized recommendations. The combination of mini-program classification features and user behavior data enables the push model to accurately capture user interests and preferences and generate a highly relevant recommendation list. This not only improves user experience and satisfaction, but also promotes the use and promotion of mini-programs.

[0088] For example, suppose the user is a sports enthusiast who loves fitness. He browses a number of mini-programs related to health management and sports records on the mini-program platform and frequently uses several of them. The system first generates classification features of these mini-programs through the previous steps, such as "health management" and "sports records". Then, the system uses these classification features to train the mini-program push model through the back propagation algorithm. Then, the system collects the user's behavioral data, such as browsing history and usage, and inputs this data into the push model. By analyzing the user's behavioral data and the mini-program classification features, the model generates a recommendation list and recommends more mini-programs related to health and sports to the user, such as "fitness plan" and "diet management". Through these steps, the system can efficiently and accurately recommend mini-programs that the user may be interested in, improving the user experience and the usage rate of the mini-programs.

[0089] In summary, step S103 solves the personalization and accuracy issues of mini-program recommendations by combining mini-program classification features and back-propagation algorithms to train and execute the mini-program push model. The system captures user interests and preferences through deep neural networks and back-propagation algorithms, generates highly relevant recommendation lists, and improves user satisfaction and mini-program promotion effects.

[0090] From the above description, it can be seen that the mini-program intelligent recommendation method based on functional summary provided in the embodiment of the present application can segment the text description and code data of the mini-program by using preset text analysis and pattern recognition algorithms to obtain segmented data, extract keywords and functional features from the segmented data by a preset natural language processing algorithm, and determine the corresponding mini-program functional summary features based on the keywords and functional features; determine the corresponding mini-program push model based on the mini-program classification features and the back propagation algorithm, and perform the mini-program push operation based on the mini-program push model; the present application can effectively improve the accuracy of mini-program push.

[0091] In one embodiment of the mini-program intelligent recommendation method based on function summary of the present application, see Figure 2 , and can also include the following:

[0092] Step S201: performing a minority class recognition operation on the keywords and functional features according to a preset minority class recognition rule, and determining corresponding nearest neighbor mini-program functional features according to the minority class mini-program functional features obtained after the minority class recognition operation and a preset K nearest neighbor algorithm;

[0093] Step S202: Determine the corresponding mini-program function summary features according to the minority mini-program function features and the nearest neighbor mini-program function features.

[0094] Optionally, in this embodiment, in step S201, the system needs to perform minority recognition operations on keywords and functional features according to preset minority recognition rules, and determine the corresponding nearest neighbor mini-program functional features based on the identified minority mini-program functional features and the preset K nearest neighbor algorithm. Subsequently, in step S202, the system determines the corresponding mini-program functional summary features based on the minority mini-program functional features and the nearest neighbor mini-program functional features. The following are the specific technical processes, technical principles, technical problems solved, and technical effects achieved:

[0095] First, the system needs to perform minority class recognition on the input keywords and functional features according to the preset minority class recognition rules. The main task of minority class recognition is to detect and identify features or samples that are small in number but may be of great significance in the data set. Minority class recognition rules can be implemented based on statistical methods (such as frequency analysis, anomaly detection) or machine learning algorithms (such as isolation forests, support vector machines, etc.). Through these rules, the system can effectively identify features or samples that are small in number in the data set but have an important impact on model training.

[0096] After completing the minority class identification operation, the system will obtain the minority class applet functional features. Next, the system needs to use the preset K-Nearest Neighbors (KNN) algorithm to determine the corresponding nearest neighbor applet functional features. The KNN algorithm is a commonly used supervised learning algorithm that calculates the distance between samples (such as Euclidean distance, Manhattan distance, etc.) to find the K neighbors closest to the target sample, and classify or regress based on the characteristics of these neighbors. In this step, the system will take the minority class applet functional features as the target sample and find the K neighbors closest to it to determine the nearest neighbor applet functional features.

[0097] In terms of technical principles, step S201 utilizes the basic principles of minority class identification and KNN algorithm, by identifying the less numerous but important features in the data set and inferring through the features of neighboring samples, ensuring that the model can fully utilize the minority class features and improve the overall prediction performance and robustness. The application of minority class identification rules enables the system to effectively detect and utilize those features that are uncommon but important in the data set; the use of KNN algorithm enables the system to make reasonable inferences through the features of neighboring samples and improve the accuracy of the model.

[0098] In terms of the technical problems solved, step S201 solves the problem of how to identify and utilize minority class features in complex data sets. Minority class features are often small in number in a data set, but may have important predictive value. Through minority class identification and KNN algorithm, the system can effectively detect and utilize these features, thereby improving the overall performance of the model and prediction accuracy.

[0099] In terms of the technical effects achieved, through step S201, the system can effectively identify the fewer but important features in the data set, and reasonably infer the nearest neighbor features corresponding to these features through the KNN algorithm. This not only improves the prediction performance and robustness of the model, but also ensures that the minority class features are fully utilized in the model training, avoiding the model's neglect of the minority class features.

[0100] Next, in step S202, the system needs to determine the corresponding mini-program function summary features based on the minority mini-program function features and the nearest neighbor mini-program function features. Specifically, the system will conduct a comprehensive analysis of the minority mini-program function features and the nearest neighbor mini-program function features to extract summary features that can fully describe the mini-program function. These summary features not only contain the information of the minority features, but also combine the overall features of the nearest neighbor features to ensure the comprehensiveness and accuracy of the function description.

[0101] In terms of technical principles, step S202 utilizes the basic principles of feature synthesis and extraction, and extracts comprehensive mini-program function summary features by comprehensively analyzing minority class features and nearest neighbor features. The feature synthesis process enables the system to fully utilize the information of minority class features and nearest neighbor features to improve the accuracy and comprehensiveness of function descriptions.

[0102] In terms of the technical problems to be solved, step S202 solves the problem of how to extract comprehensive and accurate mini-program function summary features based on the combination of minority class features and nearest neighbor features. Minority class features and nearest neighbor features each have important information. Through comprehensive analysis and extraction, the system can ensure the comprehensiveness and accuracy of the function description.

[0103] In terms of the technical effects achieved, through step S202, the system can extract comprehensive and accurate mini-program function summary features. This not only improves the accuracy and comprehensiveness of the function description, but also provides high-quality feature input for subsequent model training and push, improving the overall model performance and push effect.

[0104] For example, in a mini-program recommendation system, the user's behavior data and the mini-program's functional description are used as input features. Through step S201, the system can identify those features that are small in number but important in the user's behavior data (such as special features preferred by certain users), and find the mini-program functional features that are closest to these features through the KNN algorithm. Then in step S202, the system conducts a comprehensive analysis of these minority features and nearest neighbor features to extract comprehensive mini-program functional summary features. These summary features not only include the user's special preferences, but also combine the overall mini-program functional features to ensure the accuracy and personalization of the recommendation.

[0105] In another example, an online education platform wants to recommend suitable learning resources to users. Through step S201, the system can identify a small number of important features in the user's learning data (such as special learning resources preferred by certain users), and find the learning resource functional features closest to these features through the KNN algorithm. Then in step S202, the system conducts a comprehensive analysis of these minority class features and nearest neighbor features to extract comprehensive learning resource functional summary features. These summary features not only include the user's special learning preferences, but also combine the overall learning resource functional features to ensure the accuracy and personalization of the recommendation.

[0106] In general, through steps S201 and S202, the system can effectively identify and utilize minority class features, and extract comprehensive mini-program function summary features through comprehensive analysis. This process not only improves the accuracy and comprehensiveness of feature description, but also provides high-quality feature input for subsequent model training and push, improving the overall model performance and push effect.

[0107] In one embodiment of the mini-program intelligent recommendation method based on function summary of the present application, see Figure 3 , and can also include the following:

[0108] Step S301: performing a distance calculation operation on the minority mini-program functional features and the nearest neighbor mini-program functional features to determine the corresponding mini-program functional feature distances;

[0109] Step S302: Determine the corresponding synthesized mini-program functional features according to the mini-program functional feature distance and a preset ratio rule, perform feature adding operations on the mini-program functional features according to the synthesized mini-program functional features, and determine the corresponding mini-program functional summary features.

[0110] Optionally, in this embodiment, in step S301, the system needs to perform a distance calculation operation on the minority mini-program function features and the nearest neighbor mini-program function features to determine the corresponding mini-program function feature distance. Then, in step S302, the system determines the corresponding synthetic mini-program function features based on the mini-program function feature distance and the preset ratio rule, and determines the final mini-program function summary features through feature addition operations. The specific implementation process, technical principles, problems solved, and technical effects of these steps are as follows:

[0111] First, the system performs distance calculation on the minority class mini-program functional features and the nearest neighbor mini-program functional features in step S301. The purpose of distance calculation is to quantify the similarity or difference between the minority class features and the nearest neighbor features. Common distance calculation methods include Euclidean distance, Manhattan distance, cosine similarity, etc. In this step, the system may choose Euclidean distance to calculate the distance between features, because the Euclidean distance can intuitively reflect the geometric distance between two features in multidimensional space. Through distance calculation, the system can obtain the distance value between each pair of minority class features and nearest neighbor features, and these distance values ​​will be used to synthesize new features in subsequent steps.

[0112] In terms of technical principles, distance calculation is based on the basic concept of geometric distance, and quantifies their similarity by calculating the distance between feature vectors. The smaller the distance, the more similar the features are; the larger the distance, the greater the difference in features. In this way, the system can effectively identify and quantify the similarity between minority class features and nearest neighbor features.

[0113] In terms of the technical problems solved, step S301 solves the problem of how to quantitatively measure the similarity between minority class features and nearest neighbor features. Through accurate distance calculation, the system can provide basic data for subsequent feature synthesis and addition, ensuring the rationality and accuracy of the synthesized features.

[0114] Next, in step S302, the system determines the corresponding synthesized mini-program function features based on the mini-program function feature distance and the preset proportional rule. The preset proportional rule may be a weighted average, linear interpolation or other synthesis method based on the feature distance. For example, the system can assign different weights according to the size of the distance. The smaller the distance, the greater the feature weight, and the larger the distance, the smaller the feature weight. In this way, the system can synthesize a new feature vector that integrates the information of the minority class features and the nearest neighbor features.

[0115] After the synthesized Mini Program function features are determined, the system will perform feature addition operations on the Mini Program function features. The purpose of feature addition is to combine the synthesized features with the original features to form a more comprehensive and accurate feature description. The system may use feature fusion methods to splice or weightedly merge the synthesized feature vector with the original feature vector to form the final Mini Program function summary features.

[0116] In terms of technical principles, feature synthesis and feature addition are based on the basic concept of feature fusion. Through reasonable proportion rules and fusion methods, features from different sources are integrated to form a more comprehensive and accurate feature description. In this way, the system can effectively utilize the information of minority class features and nearest neighbor features to improve the accuracy and comprehensiveness of feature description.

[0117] In terms of the technical problems to be solved, step S302 solves the problem of how to synthesize new features and add features based on the integration of minority class features and nearest neighbor features. Through reasonable synthesis and addition operations, the system can ensure the comprehensiveness and accuracy of the final feature description, and provide high-quality feature input for subsequent model training and push.

[0118] In terms of the technical effects achieved, through step S302, the system can synthesize and add new mini-program functional features to form a comprehensive and accurate mini-program functional summary feature. This not only improves the accuracy and comprehensiveness of the feature description, but also provides high-quality feature input for the subsequent recommendation system, improving the overall recommendation effect and user satisfaction.

[0119] For example, on a mini-program recommendation platform, user behavior data and mini-program function descriptions are used as input features. In step S301, the system calculates the distance between the identified minority mini-program function features and the mini-program function features of its nearest neighbors. For example, users may have preferences for certain unpopular but special mini-program functions, and the system calculates the distance between these minority function features and the common function features of their nearest neighbors. Then in step S302, the system synthesizes a new function feature vector based on these distances and preset proportional rules. The system then merges this synthesized feature vector with the original function features to form a new mini-program function summary feature. This summary feature not only contains the user's special preferences, but also combines the information of common function features to ensure the accuracy and personalization of the recommendation.

[0120] In another example, an online education platform wants to recommend suitable learning resources to users. In step S301, the system will calculate the distance between the functional features of the minority learning resources preferred by the user and the functional features of the common learning resources of its nearest neighbors. For example, some users may be interested in some unpopular but efficient learning methods, and the system will calculate the distance between these minority learning method features and the common learning method features. Then in step S302, the system will synthesize a new learning resource functional feature vector based on these distances and preset proportional rules. Then, the system merges this synthesized feature vector with the original learning resource functional features to form a new learning resource functional summary feature. This summary feature not only includes the user's special learning preferences, but also combines the characteristics of common learning resources to ensure the accuracy and personalization of the recommendation.

[0121] In general, through steps S301 and S302, the system can effectively calculate the distance between features, synthesize new features, and add features to form a comprehensive and accurate summary of the mini program's functions. This process not only improves the accuracy and comprehensiveness of feature descriptions, but also provides high-quality feature input for subsequent recommendation systems, improving the overall recommendation effect and user satisfaction.

[0122] In one embodiment of the mini-program intelligent recommendation method based on function summary of the present application, see Figure 4 , and can also include the following:

[0123] Step S401: input the conversion code into the bidirectional encoder layer of the setting information extraction model, perform a vector conversion operation on the conversion code according to a preset word embedding algorithm, and determine a corresponding encoding vector;

[0124] Step S402: performing a coding stacking operation according to the coding vector and a preset number of coding structure layers to determine corresponding semantic features.

[0125] Optionally, in this embodiment, in step S401, the system first inputs the conversion code into the bidirectional encoder layer of the setting information extraction model. The bidirectional encoder layer is a commonly used structure in deep learning models, which can capture the contextual information of the input sequence. In the specific implementation process, the system will use technologies such as bidirectional long short-term memory network (Bi-LSTM) or bidirectional transformer (BERT) to process the input conversion code bidirectionally, so as to obtain richer contextual semantic information.

[0126] Next, the system performs a vector conversion operation on the conversion code according to the preset word embedding algorithm. The word embedding algorithm is a technology that converts words into high-dimensional vectors. Commonly used word embedding algorithms include Word2Vec, GloVe, and FastText. These algorithms map words to a continuous vector space by training a large-scale corpus, so that similar words are close in the vector space. The system uses the word embedding algorithm to convert the words in the conversion code into corresponding word vectors and generate a high-dimensional encoding vector representation.

[0127] The technical principle is mainly based on the bidirectional encoder and word embedding algorithm. By processing the forward and backward information of the input sequence at the same time, the bidirectional encoder can capture richer contextual dependencies and improve the model's ability to understand the input sequence. The word embedding algorithm maps words to a high-dimensional vector space so that semantically similar words are closer in the vector space, allowing the model to better understand and process the semantic relationship between words. By combining the bidirectional encoder and the word embedding algorithm, the system can generate accurate encoding vectors, providing a basis for subsequent semantic feature extraction.

[0128] In step S402, the system performs a coding stacking operation according to the coding vector and the preset number of coding structure layers to determine the corresponding semantic features. The coding stacking operation refers to stacking multiple coding layers to extract higher-level semantic features. In the specific implementation process, the system will stack the encoders layer by layer according to the preset number of coding structure layers, input the coding vector into each layer of encoder, extract and fuse semantic information layer by layer, and finally generate a high-level semantic feature representation.

[0129] The technical principle is mainly based on deep neural networks and hierarchical coding structures. By stacking encoders layer by layer, the system can extract the semantic information of the input sequence layer by layer, thereby generating a higher-level semantic feature representation. Each layer of encoder can capture semantic information at different levels. The bottom-level encoder mainly captures local, fine-grained semantic information, while the high-level encoder can capture global, abstract semantic information. By stacking multiple layers of encoders, the system can generate a more comprehensive and accurate semantic feature representation.

[0130] This solution effectively solves the problem of semantic feature extraction of input sequences. Through the bidirectional encoder and word embedding algorithm, the system can generate accurate encoding vectors to capture the contextual dependencies of the input sequence and the semantic relationships between words. Through the encoding stacking operation, the system can extract and fuse semantic information layer by layer, generate high-level semantic feature representations, and improve the model's understanding and processing capabilities of the input sequence.

[0131] In terms of technical effects, the system achieves efficient encoding and semantic feature extraction of input sequences through bidirectional encoders and word embedding algorithms. The bidirectional encoder can capture the contextual information of the input sequence, and the word embedding algorithm can convert words into high-dimensional vector representations, thereby improving the accuracy of encoding and semantic expression capabilities. By stacking multiple layers of encoders, the system can generate high-level semantic feature representations and improve the model's understanding and processing capabilities.

[0132] For example, suppose the system needs to process a user review text, such as "This phone has very good battery life, but the camera effect is average." The system first preprocesses and transforms the review text, and then inputs the transformed code into the bidirectional encoder layer, and captures the contextual information of the review text through technologies such as Bi-LSTM or BERT. Next, the system converts the words in the transformed code into corresponding word vectors according to the preset word embedding algorithm to generate a high-dimensional encoding vector representation. Then, the system stacks the encoders layer by layer according to the preset number of encoding structure layers, inputs the encoding vectors into each layer of encoders, extracts and fuses semantic information layer by layer, and finally generates a high-level semantic feature representation, such as "good battery life" and "average camera effect." Through these steps, the system can efficiently and accurately extract and understand the semantic features of user comments, providing a basis for subsequent analysis and processing.

[0133] In summary, steps S401 and S402 achieve efficient encoding and semantic feature extraction of input sequences by combining a bidirectional encoder and a word embedding algorithm. This solution not only improves the model's ability to understand and process input sequences, but also enhances the expression and fusion of semantic features, providing reliable data support for subsequent applications and analysis.

[0134] In one embodiment of the mini-program intelligent recommendation method based on function summary of the present application, see Figure 5 , and can also include the following:

[0135] Step S501: inputting the semantic features into the bidirectional long short-term memory network layer of the setting information extraction model, performing dependency capture operation according to a preset gating mechanism, and determining a corresponding semantic feature matrix;

[0136] Step S502: Determine corresponding semantic dependency features according to the semantic feature matrix and the preset memory unit.

[0137] Optionally, in this embodiment, in step S501, the system inputs the semantic features into the bidirectional long short-term memory network (Bi-LSTM) layer of the setting information extraction model. The bidirectional long short-term memory network is a neural network structure that can capture long-term and short-term dependencies in sequence data. In the specific implementation process, the system inputs the preprocessed semantic features into the Bi-LSTM layer, and processes the input sequence through the LSTM units in the forward and backward directions to obtain more comprehensive context information.

[0138] The working principle of bidirectional LSTM is that it can consider the context information of sequence data at the same time. Each LSTM unit controls the inflow, retention and output of information through three gating mechanisms: input gate, forget gate and output gate, so as to effectively capture the dependencies in sequence data. The forward LSTM processes from the beginning to the end of the sequence, while the backward LSTM processes from the end to the beginning. Finally, the outputs of the two are spliced ​​to generate a semantic feature matrix containing rich context information.

[0139] Technically, the gating mechanism of bidirectional LSTM performs well in capturing long-term and short-term dependencies. The input gate determines how much of the current input information is stored in the memory cell, the forget gate controls the degree of retention of existing information in the memory cell, and the output gate determines which information is output from the memory cell to the next layer. These gating mechanisms enable LSTM to effectively handle long-distance dependency problems, and the bidirectional structure further enhances the model's ability to understand sequence data.

[0140] In step S502, the system determines the corresponding semantic dependency features based on the semantic feature matrix and the preset memory unit. The semantic feature matrix contains the semantic information processed by the Bi-LSTM layer, and the preset memory unit is used to further extract and store the key parts of this information. In the specific implementation process, the system will input the semantic feature matrix into the preset memory unit and extract the key semantic dependency features through specific operations.

[0141] The memory unit can be implemented using a variety of technologies, such as attention mechanism, fully connected layer, etc. The attention mechanism can dynamically adjust the importance of each part according to the correlation of different parts of the input sequence, thereby extracting key semantic dependency features. The fully connected layer fuses the information in the semantic feature matrix through weighted summation to generate a high-level semantic dependency feature representation. Regardless of the technology used, the core task of the memory unit is to extract the most important semantic dependency information for the current task from the semantic feature matrix.

[0142] In terms of technical principles, the attention mechanism calculates the attention weight of each element in the input sequence and dynamically adjusts its contribution to the final output, so as to more accurately capture the key dependencies in the sequence. The fully connected layer performs weighted summation of input features through linear transformation, which can effectively fuse and extract feature information. Both technologies help to extract key semantic dependency features from the semantic feature matrix and improve the model's understanding and processing capabilities.

[0143] This solution effectively solves the problem of capturing long-term and short-term dependencies in sequence data. Through the bidirectional LSTM layer, the system can simultaneously consider the contextual information of the sequence data and generate a semantic feature matrix containing rich contextual information. Through the preset memory unit, the system can extract and store key semantic dependency features from the semantic feature matrix, improving the model's understanding and processing capabilities of the input sequence.

[0144] In terms of technical effects, the system achieves efficient extraction of semantic features and capture of dependencies through bidirectional LSTM layers and preset memory units. The bidirectional LSTM layer can effectively capture long-term and short-term dependencies in sequence data and generate a semantic feature matrix containing rich contextual information. The preset memory unit extracts key semantic dependency features from the semantic feature matrix through the attention mechanism or fully connected layer, improving the model's understanding and processing capabilities.

[0145] For example, suppose the system needs to process a user review text, such as "This phone has very good battery life, but the camera effect is average." The system first preprocesses the review text, extracts semantic features, and inputs it into the bidirectional LSTM layer. Through the forward and backward LSTM units, the system captures the contextual information of the review text and generates a semantic feature matrix. Then, the system inputs the semantic feature matrix into the preset memory unit, and extracts key semantic dependency features such as "battery life" and "camera effect" through the attention mechanism or the fully connected layer. Through these steps, the system can efficiently and accurately extract and understand the semantic dependency features of user comments, providing a basis for subsequent analysis and processing.

[0146] In summary, steps S501 and S502 achieve efficient extraction of semantic features and capture of dependencies by combining bidirectional LSTM and preset memory units. This solution not only improves the model's ability to understand and process sequence data, but also enhances the expression and fusion of semantic features, providing reliable data support for subsequent applications and analysis.

[0147] In one embodiment of the mini-program intelligent recommendation method based on function summary of the present application, see Figure 6 , and can also include the following:

[0148] Step S601: inputting the semantic dependency feature into the conditional random field layer of the setting information extraction model to determine a label sequence corresponding to the semantic dependency feature;

[0149] Step S602: Determine the corresponding mini-program classification features according to the tag sequence and a preset Viterbi algorithm.

[0150] Optionally, in this embodiment, in step S601, the system first inputs the semantic dependency features into the conditional random field (CRF) layer of the set information extraction model. Conditional random field is a probabilistic graph model, which is often used in sequence labeling tasks. Its core is to capture the label dependency in sequence data by defining a global feature function. In the specific implementation process, the system will take the semantic dependency features as input and use the CRF layer to model the dependency and mutual influence between these features, so as to determine the label sequence corresponding to these features.

[0151] The technical principle of the CRF layer is to comprehensively consider the dependency between input features and labels by defining state feature functions and transfer feature functions. The state feature function describes the relationship between the feature of a certain position and the label of that position, while the transfer feature function describes the relationship between adjacent labels. By maximizing the conditional probability, the CRF layer can assign the most likely label to each position in the input sequence, thereby generating a label sequence.

[0152] The technical problem to be solved is how to make full use of the dependencies between input features in sequence labeling tasks to improve the accuracy of label sequences. Traditional sequence labeling methods often only consider local features and ignore global dependencies. CRF, through the definition of global feature functions, can capture the complex dependencies between input features and improve the accuracy and consistency of labeling results.

[0153] In step S602, the system determines the corresponding mini-program classification features based on the generated tag sequence and the preset Viterbi algorithm. The Viterbi algorithm is a dynamic programming algorithm that is often used to find the optimal state sequence in the hidden Markov model (HMM). In the specific implementation process, the system will use the Viterbi algorithm to calculate the optimal path based on the generated tag sequence to determine the mini-program classification features.

[0154] The technical principle of the Viterbi algorithm is to recursively calculate the optimal state at each moment and record the optimal path. Starting from the starting state, the algorithm gradually calculates the optimal state at each moment and connects these states to eventually form a global optimal path. In this way, the system can find the optimal mini-program classification features based on the tag sequence.

[0155] Technically, the combination of the CRF layer and the Viterbi algorithm achieves efficient conversion from semantic dependency features to mini-program classification features. The CRF layer captures the complex dependencies between input features and generates accurate label sequences through the definition of global feature functions. The Viterbi algorithm uses dynamic programming to find the optimal path to determine the corresponding mini-program classification features. The combination of the two enables the system to accurately and efficiently extract mini-program classification features from the input semantic dependency features.

[0156] Through this method, the system effectively solves the problems of sequence labeling and classification feature extraction. Traditional methods often only consider local features and ignore global dependencies, resulting in inaccurate and inconsistent labeling results. However, through the definition of the global feature function of the CRF layer, the system can capture the complex dependencies between input features and improve the accuracy of the labeling results. Through the dynamic programming of the Viterbi algorithm, the system can find the optimal path from the label sequence, determine the corresponding mini-program classification features, and improve the accuracy and efficiency of classification feature extraction.

[0157] In terms of technical effects, the system achieves efficient conversion of semantic dependency features to mini-program classification features through the CRF layer and the Viterbi algorithm. The CRF layer can capture the complex dependencies between input features and generate accurate label sequences, while the Viterbi algorithm uses dynamic programming to find the optimal path and determine the corresponding mini-program classification features. This method not only improves the accuracy of sequence labeling and classification feature extraction, but also enhances the overall processing capabilities of the system.

[0158] For example, suppose the system needs to process a user review text, such as "This phone has very good battery life, but the camera effect is average." The system first preprocesses and extracts features from the review text to generate semantically dependent features. These semantically dependent features are then input into the CRF layer, and the state feature function and the transition feature function are used to generate a label sequence, such as "positive," "neutral," "negative," and so on. Next, the system uses the Viterbi algorithm to calculate the optimal path based on the generated label sequence and determine the classification features of the mini program, such as "good battery life" corresponds to "positive," and "average camera effect" corresponds to "neutral." Through these steps, the system can efficiently and accurately extract and understand the classification features of user comments, providing a basis for subsequent analysis and processing.

[0159] In summary, steps S601 and S602 achieve efficient conversion of semantic dependency features to mini-program classification features through the combination of the CRF layer and the Viterbi algorithm. This solution not only improves the accuracy of sequence labeling and classification feature extraction, but also enhances the overall processing capability of the system, providing reliable data support for subsequent applications and analysis.

[0160] In one embodiment of the mini-program intelligent recommendation method based on function summary of the present application, see Figure 7 , and can also include the following:

[0161] Step S701: performing a weight matching operation according to the decision tree prediction feature and the regression tree prediction feature output by the pruning layer in the preset regression tree model to determine the corresponding decision tree prediction feature weight and regression tree prediction feature weight;

[0162] Step S702: Perform a feature weighted average operation based on the decision tree prediction feature weights and the regression tree prediction feature weights to determine the corresponding mini-program classification features.

[0163] Optionally, another method for determining the corresponding mini-program classification features is also provided in this embodiment. In step S701, the system first performs a weight matching operation based on the decision tree prediction features and the regression tree prediction features output by the pruning layer in the preset regression tree model to determine the corresponding decision tree prediction feature weights and regression tree prediction feature weights. In the specific implementation process, the system will obtain prediction features from the decision tree and regression tree models respectively, and evaluate the importance of each feature through a certain matching algorithm, so as to assign appropriate weights to each feature.

[0164] Decision tree and regression tree models are two commonly used machine learning models, used for classification and regression tasks respectively. Decision trees divide data into different categories through a series of decision nodes, while regression trees predict numerical outputs through continuous segmentation. In practical applications, these two models can capture different features in the data and provide support for subsequent analysis and decision making. The pruning layer is an important step in the regression tree model, which aims to reduce the complexity of the model and improve the generalization ability of the model by removing unimportant nodes.

[0165] In the weight matching operation, the system needs to comprehensively consider the prediction features of the decision tree and regression tree models, and evaluate the importance of each feature through a certain algorithm. Commonly used weight matching algorithms include information gain, Gini coefficient, mean square error, etc. These algorithms can calculate the weight of each feature based on the contribution of the feature to the model output, thereby determining the prediction feature weights of the decision tree and the regression tree.

[0166] The technical problem to be solved is mainly how to effectively integrate the prediction features of decision tree and regression tree models and improve the accuracy of feature weights. Traditional methods often only consider the features of a single model, ignoring the complementarity between features of different models, resulting in unreasonable feature weight distribution. Through the weight matching operation, the system can comprehensively consider the prediction features of decision tree and regression tree models, evaluate the importance of each feature, and thus improve the accuracy and rationality of feature weights.

[0167] In step S702, the system performs a feature weighted average operation based on the determined decision tree prediction feature weights and regression tree prediction feature weights to determine the corresponding mini-program classification features. In the specific implementation process, the system will perform a weighted average of the prediction features of the decision tree and regression tree models according to their weights to generate a comprehensive mini-program classification feature.

[0168] The technical principle of feature weighted averaging operation is to fuse the prediction features of different models through weighted summation. Each feature contributes a corresponding value according to its weight, thus forming a comprehensive feature representation as a whole. This method can effectively combine the advantages of different models and improve the accuracy and stability of feature representation.

[0169] Technically, the combination of weight matching and weighted averaging achieves efficient conversion from multi-model prediction features to comprehensive classification features. The weight matching operation uses a certain algorithm to evaluate the importance of each model's prediction features and determine reasonable feature weights. The weighted averaging operation combines the prediction features of different models through weighted summation to generate comprehensive classification features. In this way, the system can extract more accurate and stable classification features from the prediction features of multiple models.

[0170] Through this method, the system effectively solves the problem of multi-model prediction feature integration and classification feature extraction. Traditional methods often only consider the features of a single model, ignoring the complementarity between different model features, resulting in inaccurate classification feature extraction. Through weight matching and weighted average operations, the system can comprehensively consider the prediction features of different models and improve the accuracy and stability of classification feature extraction.

[0171] In terms of technical effects, the system achieves efficient integration of multi-model prediction features and classification feature extraction through weight matching and weighted averaging operations. The weight matching operation can evaluate the importance of each model's prediction features and determine reasonable feature weights, while the weighted averaging operation fuses the prediction features of different models through weighted summation to generate comprehensive classification features. This method not only improves the accuracy of classification feature extraction, but also enhances the overall processing capability of the system.

[0172] For example, suppose the system needs to process a piece of user behavior data, such as the user's click, browsing, and purchase records. The system first predicts the user's behavior characteristics through decision tree and regression tree models, and removes unimportant nodes through pruning operations. Next, the system evaluates the importance of each predicted feature through weight matching operations and determines reasonable feature weights. Then, through weighted average operations, the predicted features of different models are fused to generate comprehensive user classification features, such as "high-value users" and "potential churn users". Through these steps, the system can efficiently and accurately extract and understand the classification features of user behavior, providing a basis for subsequent marketing and recommendations.

[0173] In summary, steps S701 and S702 achieve efficient integration of multi-model prediction features and classification feature extraction through weight matching and weighted average operations. This solution not only improves the accuracy of classification feature extraction, but also enhances the overall processing capability of the system, providing reliable data support for subsequent applications and analysis.

[0174] In order to effectively improve the accuracy of mini-program push, the present application provides an embodiment of a mini-program intelligent recommendation device based on a function summary for implementing all or part of the contents of the mini-program intelligent recommendation method based on a function summary, see Figure 8 The mini-program intelligent recommendation device based on function summary specifically includes the following contents:

[0175] The user portrait collection module 10 is used to collect the historical mini-program access data sent by the user, slice the text description and code data of the mini-program using a preset text analysis and pattern recognition algorithm to obtain sliced ​​data, extract keywords and functional features from the sliced ​​data using a preset natural language processing algorithm, and determine the corresponding mini-program functional summary features based on the keywords and functional features;

[0176] The mini-program classification feature module 20 is used to input the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic feature, input the semantic feature into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency feature, input the semantic dependency feature into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification feature;

[0177] The mini-program push module 30 is used to determine the corresponding mini-program push model according to the mini-program classification characteristics and the back propagation algorithm, and perform the mini-program push operation according to the mini-program push model.

[0178] From the above description, it can be seen that the mini-program intelligent recommendation device based on functional summary provided in the embodiment of the present application can segment the text description and code data of the mini-program by using preset text analysis and pattern recognition algorithms to obtain segmented data, extract keywords and functional features from the segmented data by a preset natural language processing algorithm, and determine the corresponding mini-program functional summary features based on the keywords and functional features; determine the corresponding mini-program push model based on the mini-program classification features and the back propagation algorithm, and perform the mini-program push operation based on the mini-program push model; the present application can effectively improve the accuracy of mini-program push.

[0179] From the hardware level, in order to effectively improve the accuracy of mini-program push, the present application provides an embodiment of an electronic device for implementing all or part of the mini-program intelligent recommendation method based on function profile, and the electronic device specifically includes the following contents:

[0180] Processor, memory, communication interface and bus; wherein the processor, memory and communication interface communicate with each other through the bus; the communication interface is used to realize information transmission between the small program intelligent recommendation device based on function summary and the core business system, user terminal and related database and other related equipment; the logic controller can be a desktop computer, a tablet computer and a mobile terminal, etc., but the present embodiment is not limited thereto. In the present embodiment, the logic controller can be implemented with reference to the embodiment of the small program intelligent recommendation method based on function summary and the embodiment of the small program intelligent recommendation device based on function summary in the embodiment, and the contents thereof are incorporated herein and the repeated parts are not repeated.

[0181] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0182] In actual applications, part of the mini-program intelligent recommendation method based on the function summary can be executed on the electronic device side as described above, or all operations can be completed in the client device. The specific selection can be based on the processing capability of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor.

[0183] The client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and other implementation scenarios may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0184] Fig. 9 FIG. 9 is a schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Fig. 9 As shown, the electronic device 9600 may include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that Fig. 9 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0185] In one embodiment, the function of the mini-program intelligent recommendation method based on the function summary can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0186] Step S101: using a preset text analysis and pattern recognition algorithm to segment the text description and code data of the mini program to obtain segmented data, extracting keywords and functional features from the segmented data using a preset natural language processing algorithm, and determining corresponding mini program functional summary features based on the keywords and functional features;

[0187] Step S102: Input the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic feature, input the semantic feature into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency feature, input the semantic dependency feature into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification feature;

[0188] Step S103: Determine a corresponding mini-program push model according to the mini-program classification features and the back propagation algorithm, and perform a mini-program push operation according to the mini-program push model.

[0189] From the above description, it can be seen that the electronic device provided in the embodiment of the present application adopts a preset text analysis and pattern recognition algorithm to segment the text description and code data of the mini-program to obtain segmented data, extracts keywords and functional features from the segmented data through a preset natural language processing algorithm, and determines the corresponding mini-program functional summary features based on the keywords and functional features; determines the corresponding mini-program push model based on the mini-program classification features and the back propagation algorithm, and performs the mini-program push operation based on the mini-program push model; the present application can effectively improve the accuracy of mini-program push.

[0190] In another embodiment, the mini-program intelligent recommendation device based on the functional summary can be configured separately from the central processing unit 9100. For example, the mini-program intelligent recommendation device based on the functional summary can be configured as a chip connected to the central processing unit 9100, and the function of the mini-program intelligent recommendation method based on the functional summary can be implemented through the control of the central processing unit.

[0191] like Fig. 9 As shown, the electronic device 9600 may also include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Fig. 9 In addition, the electronic device 9600 may also include Fig. 9 For components not shown, reference may be made to the prior art.

[0192] like Fig. 9 As shown, the central processing unit 9100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0193] The memory 9140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 9100 may execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0194] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0195] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 9600 through the central processor 9100.

[0196] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0197] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0198] Based on different communication technologies, multiple communication modules 9110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module, etc. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby realizing a common telecommunication function. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is also coupled to the central processor 9100, so that recording can be performed on the local machine through the microphone 9132, and the sound stored on the local machine can be played through the speaker 9131.

[0199] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all the steps of the method for intelligently recommending small programs based on function summaries in the above embodiments, where the execution subject is a server or a client. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, all the steps of the method for intelligently recommending small programs based on function summaries in the above embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0200] Step S101: using a preset text analysis and pattern recognition algorithm to segment the text description and code data of the mini program to obtain segmented data, extracting keywords and functional features from the segmented data using a preset natural language processing algorithm, and determining corresponding mini program functional summary features based on the keywords and functional features;

[0201] Step S102: Input the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic feature, input the semantic feature into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency feature, input the semantic dependency feature into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification feature;

[0202] Step S103: Determine a corresponding mini-program push model according to the mini-program classification features and the back propagation algorithm, and perform a mini-program push operation according to the mini-program push model.

[0203] From the above description, it can be seen that the computer-readable storage medium provided in the embodiment of the present application adopts a preset text analysis and pattern recognition algorithm to segment the text description and code data of the mini-program to obtain segmented data, extracts keywords and functional features from the segmented data through a preset natural language processing algorithm, and determines the corresponding mini-program functional summary features based on the keywords and functional features; determines the corresponding mini-program push model based on the mini-program classification features and the back propagation algorithm, and executes the mini-program push operation based on the mini-program push model; the present application can effectively improve the accuracy of mini-program push.

[0204] The embodiments of the present application also provide a computer program product capable of implementing all the steps of the method for intelligently recommending small programs based on function summaries in the above embodiments, where the execution subject is a server or a client. When the computer program / instruction is executed by a processor, the steps of the method for intelligently recommending small programs based on function summaries are implemented. For example, the computer program / instruction implements the following steps:

[0205] Step S101: using a preset text analysis and pattern recognition algorithm to segment the text description and code data of the mini program to obtain segmented data, extracting keywords and functional features from the segmented data using a preset natural language processing algorithm, and determining corresponding mini program functional summary features based on the keywords and functional features;

[0206] Step S102: Input the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic feature, input the semantic feature into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency feature, input the semantic dependency feature into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification feature;

[0207] Step S103: Determine a corresponding mini-program push model according to the mini-program classification features and the back propagation algorithm, and perform a mini-program push operation according to the mini-program push model.

[0208] From the above description, it can be seen that the computer program product provided in the embodiment of the present application adopts a preset text analysis and pattern recognition algorithm to segment the text description and code data of the mini-program to obtain segmented data, extracts keywords and functional features from the segmented data through a preset natural language processing algorithm, and determines the corresponding mini-program functional summary features based on the keywords and functional features; determines the corresponding mini-program push model based on the mini-program classification features and the back propagation algorithm, and executes the mini-program push operation based on the mini-program push model; the present application can effectively improve the accuracy of mini-program push.

[0209] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0210] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0211] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0212] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0213] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for intelligently recommending mini-programs based on function summaries, characterized in that: The method comprises: Using a preset text analysis and pattern recognition algorithm to slice the text description and code data of the mini program to obtain sliced ​​data, extracting keywords and functional features from the sliced ​​data using a preset natural language processing algorithm, and determining corresponding mini program functional summary features based on the keywords and functional features; Input the mini-program function summary features into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic features, input the semantic features into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency features, input the semantic dependency features into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification features; The corresponding mini-program push model is determined according to the mini-program classification features and the back-propagation algorithm, and the mini-program push operation is performed according to the mini-program push model.

2. The method for intelligently recommending mini-programs based on function summaries according to claim 1, characterized in that: Determining the corresponding mini-program function summary features according to the keywords and function features includes: Performing a minority class recognition operation on the keywords and functional features according to a preset minority class recognition rule, and determining the corresponding nearest neighbor mini program functional features according to the minority class mini program functional features obtained after the minority class recognition operation and a preset K nearest neighbor algorithm; The corresponding mini-program function summary features are determined according to the minority mini-program function features and the nearest neighbor mini-program function features.

3. The method for intelligently recommending mini-programs based on function summaries according to claim 2, characterized in that: The determining of the corresponding nearest neighbor mini-program functional features according to the minority mini-program functional features obtained after the minority class identification operation and a preset K nearest neighbor algorithm includes: Performing a distance calculation operation on the minority mini-program functional features and the nearest neighbor mini-program functional features to determine the corresponding mini-program functional feature distances; The corresponding synthesized mini-program functional features are determined according to the mini-program functional feature distance and a preset ratio rule, and the mini-program functional features are subjected to feature addition operations according to the synthesized mini-program functional features to determine the corresponding mini-program functional summary features.

4. The method for intelligently recommending mini-programs based on function summaries according to claim 1, characterized in that: The step of inputting the conversion code into the bidirectional encoder layer of the setting information extraction model to perform a semantic conversion operation and determine the corresponding semantic features includes: Input the conversion code into the bidirectional encoder layer of the setting information extraction model, perform a vector conversion operation on the conversion code according to a preset word embedding algorithm, and determine a corresponding encoding vector; A coding stacking operation is performed according to the coding vector and a preset number of coding structure layers to determine corresponding semantic features.

5. The method for intelligently recommending mini-programs based on function summaries according to claim 1, characterized in that: The step of inputting the semantic features into the bidirectional long short-term memory network layer of the setting information extraction model to perform a dependency capture operation and determining corresponding semantic dependency features includes: Inputting the semantic features into the bidirectional long short-term memory network layer of the setting information extraction model, performing dependency capture operations according to a preset gating mechanism, and determining a corresponding semantic feature matrix; The corresponding semantic dependency features are determined according to the semantic feature matrix and the preset memory unit.

6. The method for intelligently recommending mini-programs based on function summaries according to claim 1, characterized in that: The step of inputting the semantic dependency feature into the conditional random field layer of the setting information extraction model to determine the corresponding mini-program classification feature includes: Inputting the semantic dependency feature into the conditional random field layer of the setting information extraction model to determine a label sequence corresponding to the semantic dependency feature; The corresponding mini-program classification features are determined according to the tag sequence and a preset Viterbi algorithm.

7. The method for intelligently recommending mini-programs based on function summaries according to claim 1, characterized in that: The determining of the corresponding mini program classification features also includes: Perform a weight matching operation based on the decision tree prediction features and the regression tree prediction features output by the pruning layer in the preset regression tree model to determine the corresponding decision tree prediction feature weights and regression tree prediction feature weights; A feature weighted averaging operation is performed based on the decision tree prediction feature weights and the regression tree prediction feature weights to determine the corresponding mini-program classification features.

8. A device for intelligently recommending small programs based on function summaries, characterized in that: The device comprises: A user portrait collection module is used to collect historical mini-program access data sent by users, slice the text description and code data of the mini-program using a preset text analysis and pattern recognition algorithm to obtain sliced ​​data, extract keywords and functional features from the sliced ​​data using a preset natural language processing algorithm, and determine the corresponding mini-program functional summary features based on the keywords and functional features; A mini-program classification feature module is used to input the mini-program function summary feature into the input layer of the setting information extraction model for encoding conversion operation, determine the corresponding conversion code, input the conversion code into the bidirectional encoder layer of the setting information extraction model for semantic conversion operation, determine the corresponding semantic feature, input the semantic feature into the bidirectional long short-term memory network layer of the setting information extraction model for dependency capture operation, determine the corresponding semantic dependency feature, input the semantic dependency feature into the conditional random field layer of the setting information extraction model, and determine the corresponding mini-program classification feature; The mini-program push module is used to determine the corresponding mini-program push model according to the mini-program classification characteristics and the back propagation algorithm, and perform the mini-program push operation according to the mini-program push model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the mini-program intelligent recommendation method based on functional summary as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligently recommending mini-programs based on functional summaries as described in any one of claims 1 to 7 are implemented.

Images