CN-122022958-A - Application service matching method, device, computer equipment and storage medium

Abstract

The embodiment of the application provides an application service matching method, a device, computer equipment and a storage medium, wherein the method comprises the steps of collecting user data, analyzing the user data and constructing a user portrait, wherein the user data comprises user types, technical requirements, behavior attributes, preference attributes and user service fields; the method comprises the steps of collecting various application service data as service resources, classifying the service resources, constructing service portraits, wherein the application service data comprises function types, interface types, service fields, opening degrees, application types, application scene fields and application opening degrees, matching the user portraits with the service portraits based on a mixed weight distribution and depth feature cross frame, generating an application service recommendation list corresponding to a user according to a matching result, realizing more accurate service matching and recommendation, meeting personalized requirements of the user, and improving service quality and user experience.

Inventors

• Nie Zhimi
• XIA YUE
• ZHANG YANG

Assignees

• 北京比格大数据有限公司

Dates

Publication Date

20260512

Application Date

20260213

Claims (10)

1. 1. An application service matching method, comprising: ‌ collecting user data, analyzing the user data, and constructing a user portrait, wherein the user data comprises user types, technical requirements, behavior attributes, preference attributes and user service fields; Collecting various application service data as service resources, classifying the service resources, and constructing a service portrait, wherein the application service data comprises a function type, an interface type, a service field, an opening degree, an application type, an application scene field and an application opening degree; and matching the user portrait with the service portrait based on the mixed weight distribution and depth feature cross frame, and generating an application service recommendation list corresponding to the user according to a matching result.
2. 2. The method of claim 1, wherein the hybrid weight distribution and depth feature intersection framework comprises a high-to-low weight base matching layer, a context adaptation layer, a dynamic preference layer and a real-time feedback layer, wherein the base matching layer is a preset weighted cosine similarity calculation function ‌, the context adaptation layer is a preset Sigmoid function, the dynamic preference layer is a preset factorizer ‌, and the real-time feedback layer is a preset long-short-term memory network ‌ model.
3. 3. The method of claim 2, wherein the predetermined weighted cosine similarity calculation function is the following expression: ; wherein S base is the composite score of the base matching layer, wi is the weight of the ith factor, For functional demand vector And performance index vector Cosine similarity between the two, n is the total number of factors, The preset Sigmoid function is the following expression: ; Wherein S context is the score of the context adaptation layer, For the degree of matching of the network quality, In order to be a device compatibility index, As a weight coefficient of the quality of the network, For the device compatibility weighting factor to be a factor, The preset factorizer ‌ is the following expression: ; Wherein, the For the score of the dynamic preference layer, As a global bias term, As a linear weight of the i-th feature, For the value of the i-th feature, Is the hidden vector of the i-th feature, Is the hidden vector of the j-th feature, Is the dot product of the hidden vectors of feature i and feature j, As a total number of features, The preset long-term memory network ‌ model is the following expression: ; Wherein, the For the hidden state vector of time step t, For the hidden state vector of time step t-1, As an input vector for the time step t, For a hidden state to hidden state weight matrix, For the weight matrix to be input into the hidden state, The function is activated for sigmoid.
4. 4. The method according to claim 2, wherein the method further comprises: collecting the latest N times of user behavior data of a designated user, and determining the latest N times of user behavior characteristics according to the latest N times of user behavior data; Determining a user portrait according to the latest N times of user behavior characteristics, and matching application services corresponding to the user portrait; based on feedback data of the user to the application service, weight distribution of a basic matching layer, a context adapting layer, a dynamic preference layer and a real-time feedback layer is adjusted.
5. 5. The method of claim 1, wherein analyzing the user data to construct a user representation comprises: extracting the characteristics of the known user data to obtain user attribute characteristics and user behavior characteristics for training; Selecting a corresponding user portrait prediction model for modeling according to the requirements and application scenes of the user portraits; And training the user portrait prediction model by using the user attribute features and the user behavior features for training, and predicting and classifying new user data based on the trained user portrait prediction model to generate a user portrait.
6. 6. The method of claim 5, wherein the method further comprises: Updating the user portrait in real time according to the new behavior data of the user; And establishing a user feedback mechanism, and adjusting parameters of the user portrait prediction model according to feedback of the user on the recommended application service.
7. 7. The method according to claim 1, wherein the method further comprises: acquiring user behavior data and application service use data of a designated user in real time, wherein the application service use data comprises service calling success rate and user residence time; dynamically updating the user portraits according to user behavior data of a designated user and application service usage data, constructing domain migration features under the condition that domains between the updated user portraits and the user portraits before updating are different, triggering domain switching alarms, and matching the updated user portraits with corresponding application services.
8. 8. An application service matching apparatus, comprising: The first construction module is used for collecting user data ‌, analyzing the user data and constructing a user portrait, wherein the user data comprises user types, technical requirements, behavior attributes, preference attributes and user service fields; The second construction module is used for ‌ collecting various application service data as service resources, classifying the service resources and constructing service portraits, wherein the application service data comprises a function type, an interface type, a business field, an opening degree, an application type, an application scene field and an application opening degree; And the matching module is used for ‌ matching the user portrait with the service portrait based on the mixed weight distribution and depth feature cross frame, and generating an application service recommendation list corresponding to the user according to a matching result.
9. 9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the application service matching method of any of claims 1 to 7 when the computer program is executed.
10. 10. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the application service matching method of any of claims 1 to 7.

Description

Application service matching method, device, computer equipment and storage medium Technical Field The present application relates to the field of artificial intelligence technologies, and in particular, to an application service matching method, an apparatus, a computer device, and a storage medium. Background With the rapid growth of the internet, users' demands for services are increasingly tending toward personalization and diversification. The service resources facing users are increasingly abundant, but how to quickly and accurately find the service meeting the self requirements becomes a difficult problem, the original service matching and recommending technology often depends on simple rule matching or statistical models based on user behaviors, the methods often have poor effects in complex and changeable actual scenes, accurate service matching and recommending are difficult to realize, and the requirements of users on service accuracy and individuation are difficult to meet. In the related art, collaborative filtering recommendation systems find other users with interests similar to those of a target user by analyzing similarities between users, and recommend services or applications that the users like. This approach does not require detailed descriptions of the service or application itself, but only relies on the user's behavioral data. The similarity between users or items is emphasized and recommendations are made using the user data. But there is a cold start problem in that the recommending effect is limited when the new user or the new service lacks sufficient data. The content recommendation system makes recommendations based on characteristics of the service or application. The system will analyze the services or applications that the user liked in the past and recommend similar services or applications based on the characteristics (e.g., keywords, categories, etc.) of those services or applications. More accurate recommendation can be provided, and the method is particularly suitable for the situation that user preference is clear. But for scenes where the user preference is ambiguous or changes rapidly, the recommendation effect may be limited. The hybrid recommendation system combines the advantages of collaborative filtering and content recommendation, and comprehensively considers the characteristics of user behaviors and services or applications to conduct recommendation. Through complex algorithms and models, more comprehensive and accurate recommendation results are provided. The limitation of a single recommendation method is overcome, the accuracy and the individuation degree of recommendation are improved, but higher computing resources and algorithm complexity are required. However, the above recommendation system has the problems that firstly, the accuracy is insufficient, the conventional service matching and recommendation system often depends on a simple rule matching or a recommendation algorithm based on statistics, and the methods often have difficulty in achieving high accuracy when processing complex and changeable user requirements and service scenes. Secondly, the real-time performance is poor, and the traditional system cannot respond in time when facing to the user requirements and service scenes which change in real time. Third, cold start problems-conventional systems often have difficulty making efficient recommendations for new users or new services. Disclosure of Invention The embodiment of the application provides an application service matching method, an application service matching device, computer equipment and a storage medium. In a first aspect of the embodiment of the present application, there is provided an application service matching method, including: ‌ collecting user data, analyzing the user data, and constructing a user portrait, wherein the user data comprises user types, technical requirements, behavior attributes, preference attributes and user service fields; Collecting various application service data as service resources, classifying the service resources, and constructing a service portrait, wherein the application service data comprises a function type, an interface type, a service field, an opening degree, an application type, an application scene field and an application opening degree; and matching the user portrait with the service portrait based on the mixed weight distribution and depth feature cross frame, and generating an application service recommendation list corresponding to the user according to a matching result. In an optional embodiment of the present application, the hybrid weight distribution and depth feature cross framework includes a base matching layer, a context adaptation layer, a dynamic preference layer and a real-time feedback layer, where the base matching layer is a preset weighted cosine similarity calculation function ‌, the context adaptation layer is a preset Sigmoid function, the dynamic preference l