Information processing method and device, computer equipment, storage medium and program product

Abstract

The embodiment of the invention provides an information processing method and device, computer equipment, a storage medium and a program product. The information processing method comprises the following steps: acquiring plug-in demand information of a business object, and acquiring plug-in documents of a plurality of candidate plug-ins; obtaining a plug-in matching strategy, wherein the plug-in matching strategy is used for indicating a semantic matching process between the plug-in demand information and the plug-in document; according to a semantic matching process indicated by the plug-in matching strategy, performing semantic matching between the plug-in demand information and plug-in documents of the plurality of candidate plug-ins, and selecting a first recommended plug-in from the plurality of candidate plug-ins; determining first recommendation content related to plug-in demand information semantics in a plug-in document of the first recommendation plug-in; and performing plug-in recommendation processing on the business object according to the first recommendation plug-in and the first recommendation content. By adopting the embodiment of the invention, the plug-in recommendation accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to an information processing method, an information processing apparatus, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In the field of computer technology, with the rapid development of various software platforms and application ecosystems, the plug-in market has also flourished. The plug-in market is an online platform for providing plug-ins, from which business objects can obtain plug-ins. To improve the efficiency of plug-in selection in the plug-in market, recommendation systems have been introduced to the plug-in market to recommend plug-ins to business objects. However, current recommendation systems in the plug-in market typically recommend plug-ins based on the business object's historical plug-in acquisition information or real-time popular plug-in fields. The application of this recommendation logic in plug-in recommendation still faces some challenges. In particular, the recommended plug-ins often do not meet the actual needs of the business object, resulting in low plug-in recommendation accuracy. Therefore, how to improve the accuracy of plug-in recommendations has become a current research hotspot. Summary of the Invention

[0003] The embodiments of the present application provide an information processing method, apparatus, computer device, storage medium, and program product, which can improve the accuracy of plug-in recommendations.

[0004] In one aspect, an embodiment of the present application provides an information processing method, the information processing method comprising:

[0005] Obtain plug-in requirement information for business objects and plug-in documents for multiple candidate plug-ins;

[0006] Get the plug-in matching strategy, which is used to indicate the semantic matching process between the plug-in requirement information and the plug-in document;

[0007] Perform semantic matching between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and select a first recommended plug-in from the plurality of candidate plug-ins;

[0008] Determining, in a plug-in document of a first recommended plug-in, first recommended content semantically related to the plug-in requirement information;

[0009] Perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

[0010] Accordingly, an embodiment of the present application provides an information processing device, comprising:

[0011] An acquisition unit, used to acquire plug-in requirement information of a business object and to acquire plug-in documents of multiple candidate plug-ins;

[0012] The acquisition unit is further used to acquire the plug-in matching strategy, which is used to indicate the semantic matching process between the plug-in requirement information and the plug-in document;

[0013] a processing unit, configured to perform semantic matching between the plug-in requirement information and plug-in documents of a plurality of candidate plug-ins according to a semantic matching process indicated by the plug-in matching strategy, and select a first recommended plug-in from the plurality of candidate plug-ins;

[0014] The processing unit is further configured to determine, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information;

[0015] The processing unit is further configured to perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

[0016] In one implementation, a processing unit is configured to perform semantic matching between plug-in requirement information and plug-in documents of multiple candidate plug-ins according to a semantic matching process indicated by a plug-in matching strategy, and to select a first recommended plug-in from the multiple candidate plug-ins by specifically executing the following steps:

[0017] Based on the similarity between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins, screening out at least one plug-in document of the target plug-in from the plug-in documents of the plurality of candidate plug-ins;

[0018] According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the plug-in requirement information and the plug-in document of at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

[0019] In one implementation, a processing unit is configured to perform semantic matching between plug-in requirement information and a plug-in document of at least one target plug-in according to a semantic matching process indicated by a plug-in matching strategy, and to select a first recommended plug-in from the at least one target plug-in by performing the following steps:

[0020] Performing semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and predicting a probability of recommending at least one target plug-in to the business object;

[0021] Based on the recommendation probability of the at least one target plug-in, a first recommended plug-in is selected from the at least one target plug-in.

[0022] In one implementation, the processing unit is configured to select a first recommended plug-in from at least one target plug-in based on the recommendation probability of at least one target plug-in, and is specifically configured to perform the following steps:

[0023] Optimizing the recommendation probability of the at least one target plug-in based on the similarity between the plug-in requirement information and the plug-in document of the at least one target plug-in to obtain an optimized probability of the at least one target plug-in;

[0024] A first recommended plug-in is selected from the at least one target plug-in according to the optimization probability of the at least one target plug-in.

[0025] In one implementation, the processing unit is configured to filter out at least one target plug-in document from multiple candidate plug-in documents based on similarities between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and specifically to perform the following steps:

[0026] Get the software type of the target software for which the plug-in is to be installed;

[0027] Performing type matching processing on the software type of the target software and the plug-in service types of the multiple candidate plug-ins, and screening the plug-in documents of the multiple reference plug-ins from the plug-in documents of the multiple candidate plug-ins;

[0028] Based on the similarity between the plug-in requirement information and the plug-in documents of the multiple reference plug-ins, at least one plug-in document of the target plug-in is filtered out from the plug-in documents of the multiple reference plug-ins.

[0029] In one implementation, a processing unit is configured to perform semantic matching between plug-in requirement information and a plug-in document of at least one target plug-in according to a semantic matching process indicated by a plug-in matching strategy, and to select a first recommended plug-in from the at least one target plug-in by performing the following steps:

[0030] Adjust the plug-in requirement information according to the software description information of the target software to obtain the adjusted plug-in requirement information;

[0031] According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the adjusted plug-in requirement information and the plug-in document of at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

[0032] In one implementation, the processing unit is configured to filter out at least one target plug-in document from multiple candidate plug-in documents based on similarities between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and specifically to perform the following steps:

[0033] Get similarity adjustment factor;

[0034] Performing distribution adjustment processing on the similarities between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins according to the distribution adjustment direction corresponding to the similarity adjustment factor, thereby obtaining selection probabilities of the plug-in documents of the multiple candidate plug-ins;

[0035] Based on the selection probabilities of the plug-in documents of the multiple candidate plug-ins, at least one plug-in document of the target plug-in is screened out from the plug-in documents of the multiple candidate plug-ins.

[0036] In one implementation, a processing unit is configured to perform semantic matching between plug-in requirement information and plug-in documents of multiple candidate plug-ins according to a semantic matching process indicated by a plug-in matching strategy, and to select a first recommended plug-in from the multiple candidate plug-ins by specifically executing the following steps:

[0037] Perform semantic understanding on the plug-in requirement information to obtain the semantic features of the plug-in requirement information;

[0038] Perform semantic understanding on the plug-in document of each candidate plug-in to obtain the document semantic features corresponding to the plug-in document of each candidate plug-in;

[0039] Perform feature matching between the demand semantic features and the document semantic features corresponding to the plug-in document of each candidate plug-in, and predict the recommendation probability of each candidate plug-in to the business object;

[0040] A first recommended plug-in is selected from the multiple candidate plug-ins according to the recommendation probabilities of the multiple candidate plug-ins.

[0041] In one implementation, the processing unit is configured to, when determining, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information, specifically perform the following steps:

[0042] Obtaining a recommended content type corresponding to the plug-in requirement information, where the recommended content type is preset, or the recommended content type is obtained by performing content recognition on the plug-in requirement information;

[0043] Performing content screening processing on the plug-in document of the first recommended plug-in according to the recommended content type to obtain screened document content;

[0044] A semantic matching process is performed between the plug-in requirement information and the filtered document content, and a first recommended content semantically related to the plug-in requirement information is determined from the filtered document content.

[0045] In one implementation, the processing unit is further configured to perform the following steps:

[0046] updating the semantic matching process indicated by the plug-in matching strategy according to the recommendation feedback information to obtain an updated plug-in matching strategy;

[0047] Perform semantic matching between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins according to the semantic matching process indicated by the updated plug-in matching strategy, and select a second recommended plug-in from the multiple candidate plug-ins;

[0048] Determining, in the plug-in document of the second recommended plug-in, second recommended content semantically related to the plug-in requirement information;

[0049] Perform plug-in recommendation processing on the business object according to the second recommended plug-in and the second recommended content.

[0050] In one implementation, the training process of the information processing model includes:

[0051] Obtaining sample data for training an information processing model, the sample data including sample requirement information and plug-in documents of a plurality of sample plug-ins;

[0052] Calling the retrieval model to perform similarity matching between the sample requirement information and the plug-in documents of the multiple sample plug-ins, and obtaining the selection probability of the retrieval model for the plug-in documents of the multiple sample plug-ins;

[0053] Calling the language model to perform semantic matching between the sample requirement information and the plug-in documents of multiple sample plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and predicting the recommendation probability of the multiple sample plug-ins;

[0054] According to the difference between the selection probabilities of the plug-in documents of the multiple sample plug-ins and the recommendation probabilities of the multiple sample plug-ins, loss information of the information processing model is determined, and the retrieval model in the information processing model is trained according to the loss information.

[0055] Accordingly, an embodiment of the present application provides a computer device, comprising:

[0056] a processor suitable for implementing a computer program;

[0057] A computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the above-mentioned information processing method.

[0058] Accordingly, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is read and executed by a processor of a computer device, the computer device executes the above-mentioned information processing method.

[0059] Accordingly, an embodiment of the present application provides a computer program product, which includes a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the above-mentioned information processing method.

[0060] In an embodiment of the present application, plug-in requirement information of a business object can be obtained, and plug-in documents of multiple candidate plug-ins can be obtained, and a plug-in matching strategy can be obtained. The plug-in matching strategy can be used to indicate a semantic matching process between the plug-in requirement information and the plug-in document. According to the semantic matching process indicated by the plug-in matching strategy, semantic matching can be performed between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and a first recommended plug-in can be selected from the multiple candidate plug-ins. It can be seen that through semantic matching, the plug-in document of the selected first recommended plug-in can be semantically matched with the plug-in requirement information, that is, the plug-in document of the first recommended plug-in can better meet the needs of the business object, thereby improving the accuracy of plug-in recommendation by performing plug-in recommendation processing based on the first recommended plug-in that meets the needs of the business object. In addition, first recommended content semantically related to the plug-in requirement information can be determined in the plug-in document of the first recommended plug-in, so that the first recommended content is fully consistent with the plug-in requirement information, thereby improving the accuracy of plug-in recommendation by performing plug-in recommendation processing based on the first recommended content that meets the needs of the business object. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0062] Figure 1 This is a schematic diagram of the architecture of an information processing system provided by an embodiment of the present application;

[0063] Figure 2 This is a schematic diagram of the architecture of a server in an information processing system provided by an embodiment of the present application;

[0064] Figure 3 This is a flowchart of an information processing method provided by an embodiment of the present application;

[0065] Figure 4 This is a flowchart of another information processing method provided by an embodiment of the present application;

[0066] Figure 5aThis is a schematic diagram of a recommended scenario of a plug-in market provided by an embodiment of the present application;

[0067] Figure 5b This is a schematic diagram of another recommended scenario of the plug-in market provided in an embodiment of the present application;

[0068] Figure 5c This is a schematic diagram of another recommended scenario of the plug-in market provided in an embodiment of the present application;

[0069] Figure 6 This is a flowchart of another information processing method provided in an embodiment of the present application;

[0070] Figure 7 This is a schematic diagram summarizing an information processing method provided by an embodiment of the present application;

[0071] Figure 8 This is a schematic diagram of the structure of an information processing device provided in an embodiment of the present application;

[0072] Figure 9 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0073] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0074] The embodiment of the present application proposes an information processing method that can combine the Retrieval-augmented Generation (RAG) technology in the plug-in market to recommend plug-ins to business objects based on the needs of the business objects, aiming to recommend plug-ins that meet the needs of the business objects to the business objects through the Retrieval-augmented Generation technology. Compared with the blindness of business objects manually selecting plug-ins in the traditional plug-in market, the embodiment of the present application can provide more accurate and personalized plug-in recommendations through the Retrieval-augmented Generation technology, greatly improving the efficiency and accuracy of plug-in recommendations, as well as improving the convenience and satisfaction of business objects when selecting plug-ins.

[0075] In order to more clearly understand the technical solutions provided by the embodiments of the present application, the following first introduces the technical terms involved in the embodiments of the present application:

[0076] (1) Plug-ins:

[0077] A plug-in can be an additional component, specifically a program, used to extend or enhance the functionality of existing software. Plug-ins typically exist as independent modules that add new features, functions, or services to existing software without modifying the main program code.

[0078] (2) Plug-in market:

[0079] A plug-in market may refer to an online platform for providing plug-ins to business objects (business objects are users of the plug-in market). Providing plug-ins here refers to providing plug-in usage functions such as publishing, browsing, recommending, and downloading plug-ins. The plug-in market may include a plug-in market front-end and a plug-in market back-end; the plug-in market front-end refers to the plug-in market client used by business objects, which can be used to receive plug-in operations from business objects (for example, plug-in publishing operations, plug-in browsing operations, plug-in requirement input operations, and plug-in download operations, etc.) and provide corresponding interface interactions; the plug-in market back-end refers to the back-end server corresponding to the plug-in market client, which can be used to provide back-end technical service support for the plug-in market client.

[0080] (3) Retrieval enhancement generation technology:

[0081] Retrieval-enhanced generation technology refers to a technique that combines language models with information retrieval. In an embodiment of the present application, after obtaining the requirements of a business object, the retrieval-enhanced generation technology can retrieve plug-in information that matches the business object's requirements from the plug-in database based on the business object's requirements. The retrieval-enhanced generation technology can then use the language model to make plug-in recommendations based on this retrieved plug-in information, improving the accuracy and effectiveness of plug-in recommendations.

[0082] The retrieval enhancement generation technology can include three processes: retrieval, enhancement, and generation. Among them, retrieval refers to retrieving relevant information from an external knowledge base based on the query content, enhancement refers to embedding the query content and the retrieved relevant information into a preset prompt word template, and generation refers to inputting the retrieval-enhanced prompt words into the language model to generate the required output. In an embodiment of the present application, retrieval refers to retrieving relevant plug-in information from a plug-in database based on the needs of the business object, enhancement refers to embedding the needs of the business object and the retrieved plug-in information into a preset prompt word template, and generation refers to inputting the retrieval-enhanced prompt words into the language model, so that the language model recommends plug-ins based on the needs of the business object and the retrieved plug-in information.

[0083] The following is an introduction to the information processing system involved in the embodiments of the present application.

[0084] like Figure 1As shown, the information processing system may include a terminal device 101 and a server 102. The terminal device 101 and the server 102 may establish a direct communication connection through wired communication, or the terminal device 101 and the server 102 may establish an indirect communication connection through wireless communication, which is not limited in the present embodiment.

[0085] For the terminal device 101 : a plug-in market client of a business object may be running in the terminal device 101 .

[0086] Regarding the server 102: the server 102 may be a backend server corresponding to the plug-in market client, and may be used to provide backend technical service support to the plug-in market client. Figure 2 As shown, server 102 may include an information acquisition module 201, an information processing model 202, and a plug-in recommendation module 203. Information processing model 202 may include a retrieval model 2021 and a language model 2022 required for retrieval enhancement generation technology. Information acquisition module 201 may be used to acquire plug-in requirement information of a business object and obtain plug-in documents of multiple candidate plug-ins from a plug-in database. Retrieval model 221 may be a similarity retrieval model that performs retrieval by calculating similarity. Specifically, it may be used to select plug-in documents based on the similarity between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and select the plug-in document of at least one target plug-in. Language model 222 may be an LLM (Large Language Model) that may be used to semantically match the plug-in requirement information with the plug-in document of at least one target plug-in and predict a recommendation probability for the at least one target plug-in. Plug-in recommendation module 203 may be used to rank at least one target plug-in based on the recommendation probability of the at least one target plug-in, select a first recommended plug-in based on the ranking result, and determine first recommended content in the plug-in document of the first recommended plug-in that is semantically related to the plug-in requirement information.

[0087] In an information processing system composed of a terminal device 101 and a server 102, the plug-in recommendation process may include the following steps ①-⑧: ① The plug-in market client in the terminal device 101 can receive a plug-in requirement input operation of a business object. The content associated with the plug-in requirement input operation is plug-in requirement information, which can be used to reflect the plug-in acquisition requirements of the business object. ② The plug-in market client can send the plug-in requirement information to the server 102 through the terminal device 101. ③ The server 102 can obtain plug-in documents of multiple candidate plug-ins from the plug-in database through the information acquisition module 201. ④ The server 102 can call the retrieval model 2021 to filter out the plug-in document of at least one target plug-in that matches the plug-in requirement information from the plug-in documents of the multiple candidate plug-ins based on the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins. ⑤ The server 102 can call the language model 2022 to perform semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in, and predict the recommendation probability of at least one target plug-in. ⑥ Server 102 may use plug-in recommendation module 203 to rank the at least one target plug-in based on its recommendation probability, select a first recommended plug-in based on the ranking result, and determine first recommended content within the plug-in document of the first recommended plug-in that is semantically relevant to the plug-in requirement information. ⑦ Server 102 may send the first recommended plug-in and the first recommended content to terminal device 101. ⑧ The plug-in market client in terminal device 101 may output the first recommended plug-in and the first recommended content to the business object.

[0088] It can be seen that in Figure 1 In the information processing system shown, based on the retrieval enhancement generation technology, the plug-in document of the first recommended plug-in recommended to the business object is similar to the plug-in requirement information of the business object and semantically matches, and the first recommended content recommended to the business object is semantically related to the plug-in requirement information of the business object. In this way, the plug-in recommendation can be consistent with the needs of the business object, and can better meet the plug-in acquisition needs of the business object, thereby improving the accuracy of the plug-in recommendation.

[0089] exist Figure 1In the information processing system shown, the terminal device 101 may include, but is not limited to, any of the following: a smartphone, a tablet computer, a laptop computer, a desktop computer, a smartwatch, a smart home appliance, a smart car terminal, and an aircraft. The server 102 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It may also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms, etc. The client type of the plug-in market client may include, but is not limited to, any of the following: an application, a mini-program, or a web (World Wide Web) application, etc.

[0090] Figure 1 The information processing system of the illustrated embodiment is intended to more clearly illustrate the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided by the embodiment of the present application. Ordinary technicians in this field can know that with the evolution of the system architecture and the emergence of new business scenarios, the technical solution provided by the embodiment of the present application is also applicable to similar technical problems.

[0091] In the embodiments of this application, the collection and processing of relevant data (for example, the collection and processing of plug-in requirement information of business objects) should be strictly in accordance with the requirements of relevant laws and regulations when applied in instances, and the informed consent or separate consent of the personal information subject (or with a legal basis) should be obtained, and subsequent data use and processing should be carried out within the scope of authorization of laws and regulations and the personal information subject.

[0092] The information processing method provided in the embodiments of the present application is described in detail below.

[0093] The embodiment of the present application provides an information processing method, which includes: a method for calculating similarity, a method for filtering plug-in documents based on similarity, and a method for selecting plug-ins based on semantic matching, etc. The information processing method can be executed by a computer device, which can be, for example, a server 102 in an information processing system. Figure 3 As shown, the information processing method may include but is not limited to the following steps S301 to S305:

[0094] S301: Obtain plug-in requirement information of a business object, and obtain plug-in documents of multiple candidate plug-ins.

[0095] In step S301, the plug-in requirement information can be used to characterize the plug-in acquisition requirements of the business object, and the plug-in requirement information can be the content associated with the plug-in requirement input operation of the business object. In one implementation, the plug-in requirement input operation can be an input operation of the plug-in requirement text. For example, the plug-in requirement text can be a plug-in requirement keyword or a plug-in requirement description text. In this implementation, the plug-in requirement information can be the plug-in requirement text input by the plug-in requirement input operation of the business object. In another implementation, the plug-in requirement input operation can be an operation of selecting a plug-in type. Here, the plug-in type can be, for example, a plug-in business type (a plug-in business type refers to a business scenario type applicable to the plug-in) or a plug-in function type (a plug-in function type refers to a function type implemented by the plug-in), etc. In this implementation, the plug-in requirement information can be the plug-in type selected by the plug-in requirement input operation of the business object. It is not difficult to see that the embodiments of the present application can support a variety of plug-in requirement input operations, which can include an input operation of the plug-in requirement text or an operation of selecting a plug-in type, and can better meet a variety of plug-in requirement input requirements.

[0096] A plug-in document may refer to technical documentation written for a plug-in. The plug-in document may be used to describe the plug-in's attribute information. This attribute information may include, for example, at least one of the following: functionality, usage instructions, and interface specifications. The plug-in documents for the candidate plug-ins may be obtained from a plug-in database, which may be used to store the plug-in's installation files and plug-in documents.

[0097] S302: Obtain a plug-in matching strategy, where the plug-in matching strategy is used to indicate a semantic matching process between plug-in requirement information and plug-in documents.

[0098] In step S302, the semantic matching process refers to the steps involved in semantically matching the plugin requirement information with the plugin documents of multiple candidate plugins, as well as the order in which these steps are performed. Semantic matching is performed by a language model, and the plugin matching strategy can guide the language model in considering each step of the semantic matching process. For example, the semantic matching process may include: 1. Considering the semantics expressed in the plugin requirement information and extracting requirement semantic features from it; 2. Considering the semantics expressed in the plugin document and extracting document semantic features from it; 3. Considering the degree of feature matching between the requirement semantic features and the document semantic features; and 4. Outputting a predicted probability.

[0099] S303 : performing semantic matching between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and selecting a first recommended plug-in from the plurality of candidate plug-ins.

[0100] In step S303, in order to improve the efficiency of semantic matching, plug-in documents of a portion of candidate plug-ins (which may be referred to as target plug-ins) that are similar to the plug-in requirement information may be screened out from the plug-in documents of multiple candidate plug-ins, and semantic matching may be performed between the plug-in requirement information and the plug-in documents of the screened target plug-ins. Specifically, the following sub-steps s11 to s12 may be included:

[0101] s11, based on the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins, screen out at least one plug-in document of a target plug-in from the plug-in documents of the multiple candidate plug-ins.

[0102] In sub-step s11, the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins is calculated in any of the following ways:

[0103] The first one is to perform similarity matching between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins.

[0104] In the first similarity calculation method, similarity matching can be performed between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins to obtain the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins. Similarity matching here refers to vector similarity matching. The plug-in requirement information and plug-in documents can be converted into corresponding vectors (vectors can also be called embedding representations) and then the vector similarity can be calculated.

[0105] Specifically, the process of performing similarity matching between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins to obtain the similarity between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins may include: performing information vectorization on the plug-in requirement information to obtain a plug-in requirement vector corresponding to the plug-in requirement information; performing document vectorization on the plug-in documents of multiple candidate plug-ins to obtain document vectors corresponding to the plug-in documents of multiple candidate plug-ins; and determining the vector similarity between the plug-in requirement vector and the document vectors corresponding to the multiple candidate plug-ins as the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins.

[0106] In more detail, step S302 may be performed by a retrieval model in the information processing model. The retrieval model may adopt a dense retrieval method based on embedding representation (i.e., vector, embedding), such as a dual encoder model, to calculate the similarity between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins. It can be understood that the retrieval model may include a dual encoder model. The dual encoder model may include a query encoder and a document encoder; the query encoder may be used to perform information vectorization, where information vectorization refers to information encoding, i.e., the query encoder may be used to encode the plug-in requirement information into a plug-in requirement vector of a first length; the document encoder may be used to perform document vectorization, where document vectorization refers to document encoding, i.e., the document encoder may be used to encode the plug-in document of each candidate plug-in into a document vector of a second length; the first length and the second length may be the same or different, and the embodiment of the present application does not limit this.

[0107] The first method of calculating similarity is shown in the following formula 1:

[0108] s(d i ,x)=cos(E(d i ),E(x)) Formula 1

[0109] In the above formula 1, d i represents the plug-in document of any candidate plug-in (i-th candidate plug-in) in the first plug-in document set D. The first plug-in document set D is a set consisting of plug-in documents of multiple candidate plug-ins. D can be expressed as D = {d1, d2, ..., d k}, k is an integer greater than 1, i is a positive integer less than or equal to k; s(d i ,x) represents the similarity between the plug-in document of the i-th candidate plug-in and the plug-in requirement information; E(d i ) represents the embedded representation of the plug-in document of the i-th candidate plug-in (i.e., the document vector corresponding to the plug-in document of the i-th candidate plug-in); e(x) represents the embedded representation of the plug-in requirement information (i.e., the plug-in requirement vector corresponding to the plug-in requirement information); cos is the cosine similarity metric.

[0110] The second method is to use the plug-in document of the plug-in that has been historically obtained by the business object (which can be expressed as a historically obtained plug-in) as a reference and perform similarity matching between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins.

[0111] In a second similarity calculation method, historical plug-in acquisition information for a business object can be obtained. This historical plug-in acquisition information can include plug-in documents for historically acquired plug-ins for the business object. Using the plug-in documents for the historically acquired plug-ins as a reference, similarity matching is performed between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins to determine the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins. Similarity matching here refers to vector similarity matching. The plug-in requirement information and the plug-in documents for the historically acquired plug-ins can be fused to form a corresponding fused vector. After the plug-in documents are converted to corresponding vectors, vector similarity is calculated.

[0112] Specifically, with the plug-in document of the historically acquired plug-in as a reference, similarity matching is performed between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins to obtain the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins. The process may include: information vectorization of the plug-in requirement information to obtain a plug-in requirement vector corresponding to the plug-in requirement information; document vectorization of the plug-in document of the historically acquired plug-in to obtain a document vector corresponding to the plug-in document of the historically acquired plug-in; vector fusion processing of the plug-in requirement vector and the document vector corresponding to the historically acquired plug-in to obtain a fusion vector. The fusion processing here refers to vector addition or vector multiplication of the plug-in requirement vector and the document vector corresponding to the historically acquired plug-in, which is not limited in this embodiment of the present application; document vectorization of the plug-in documents of multiple candidate plug-ins to obtain document vectors corresponding to the plug-in documents of multiple candidate plug-ins; the vector similarity between the fusion vector and the document vectors corresponding to the multiple candidate plug-ins can be determined as the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins.

[0113] Similar to the first similarity calculation method, the similarity calculation can be performed by a retrieval model in the information processing model, which can include a dual encoder model. The dual encoder model can include a query encoder and a document encoder; the query encoder can be used to perform information vectorization, where information vectorization refers to information encoding, i.e., the query encoder can be used to encode plug-in requirement information into a plug-in requirement vector of a first length; the document encoder can be used to perform document vectorization, where document vectorization refers to document encoding, i.e., the document encoder can be used to encode the plug-in document of each candidate plug-in into a document vector of a second length, and can be used to encode the plug-in document of the historically acquired plug-in into a document vector of a second length.

[0114] Based on the introduction to the second similarity calculation method above, it can be seen that the impact of the business object's historical plug-in acquisition on the similarity is taken into account during the similarity calculation process. This is because the business object's historical plug-in acquisition can reflect the business object's plug-in acquisition preference, which can improve the accuracy of the similarity calculation.

[0115] In sub-step s11, the process of selecting at least one target plug-in document from the plug-in documents of the multiple candidate plug-ins based on the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins may include: comparing the similarity with a similarity threshold (the similarity threshold may be set based on an empirical value), and determining the plug-in documents of the candidate plug-ins whose corresponding similarity is greater than or equal to the similarity threshold as the plug-in documents of the at least one target plug-in; or sorting the plug-in documents of the multiple candidate plug-ins in descending order of similarity, and determining the plug-in documents of the candidate plug-ins ranked in the top N positions as the plug-in documents of the at least one target plug-in, where N is an integer greater than 1.

[0116] Optionally, after appropriately adjusting the similarity, plug-in documents can be screened. Specifically, a similarity adjustment factor can be obtained, and distribution adjustment processing can be performed on the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins based on the distribution adjustment direction corresponding to the similarity adjustment factor to obtain selection probabilities for the plug-in documents of the multiple candidate plug-ins. Based on the selection probabilities of the plug-in documents of the multiple candidate plug-ins, at least one plug-in document of a target plug-in can be screened from the multiple plug-in documents of the candidate plug-ins.

[0117] In detail, the process of screening out at least one target plug-in document that matches the plug-in requirement information from the plug-in documents of multiple candidate plug-ins based on the selection probability of the plug-in documents of multiple candidate plug-ins may include: comparing the selection probability with a first probability threshold, and determining the plug-in documents of the candidate plug-ins whose corresponding selection probability is greater than or equal to the first probability threshold as the plug-in documents of the at least one target plug-in; or, sorting the plug-in documents of the multiple candidate plug-ins in descending order of the selection probability, and determining the plug-in documents of the candidate plug-ins ranked in the top N positions as the plug-in documents of the at least one target plug-in, where N is an integer greater than 1.

[0118] It should be noted that the similarity adjustment factor can specifically refer to a temperature hyperparameter, which can be used to control the distribution of similarities between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins. The similarity adjustment factor (i.e., the temperature hyperparameter) can be set based on empirical values. When the similarity adjustment factor is greater than the parameter threshold, the distribution adjustment direction corresponding to the similarity adjustment factor is a distribution smoothing direction, which can be used to control the distribution of similarities between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins to be smooth (i.e., the probability distribution of the selection probability of the plug-in documents of multiple candidate plug-ins is smoothed). In addition, the larger the similarity adjustment factor, the smoother the distribution. The impact of this in the plug-in document screening process is that the number of plug-in documents of the target plug-in screened out is greater, and the diversity of target plug-ins that match the plug-in requirement information is increased. When the similarity adjustment factor is less than the parameter threshold, the distribution adjustment direction corresponding to the similarity adjustment factor is a sharp distribution direction, which can be used to control the distribution of the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins to tend to be sharp (that is, the probability distribution of the selection probability of the plug-in documents of the multiple candidate plug-ins tends to be sharp), and the smaller the similarity adjustment factor, the sharper the distribution; the impact of this in the plug-in document screening process is that the number of plug-in documents of the screened target plug-in is smaller, which increases the certainty of the target plug-in matching the plug-in requirement information.

[0119] The adjustment process of the similarity adjustment factor is shown in the following formula 2:

[0120]

[0121] In the above formula 2, P R (d i |x) represents the selection probability of any candidate plug-in (i-th candidate plug-in) in the first plug-in document set D; s(d i |x) represents the similarity between the i-th candidate plug-in and the plug-in requirement information; γ represents the similarity adjustment factor.

[0122] s12, performing semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and selecting a first recommended plug-in from the at least one target plug-in.

[0123] In sub-step s12, the process of selecting the first recommended plug-in based on semantic matching may include: performing semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and predicting the recommendation probability of recommending at least one target plug-in to the business object; and selecting a first recommended plug-in from at least one target plug-in based on the recommendation probability of at least one target plug-in.

[0124] In the above-mentioned process of selecting the first recommended plug-in based on semantic matching:

[0125] ① The process of predicting the recommendation probability based on semantic matching can be performed by the language model in the information processing model. Specifically, the plug-in documents of at least one target plug-in can form a second plug-in document set D′, and D′ can be expressed as D′={d1,d2,…,d t}, the plugin documents of each target plugin in the second plugin document set D′ (i.e., the plugin documents of at least one target plugin) can be concatenated with the plugin requirement information to obtain a set of input information, resulting in t sets of input information (i.e., at least one set of input information), where t is a positive integer less than or equal to k. These t sets of input information (i.e., at least one set of input information) can then be input into a language model, which processes the t sets of input information in parallel and ultimately outputs the recommendation probabilities corresponding to each of the t sets of input information, i.e., the recommendation probabilities of the at least one target plugin. It is not difficult to imagine that parallel processing can improve the efficiency of predicting recommendation probabilities, thereby improving the efficiency of plugin recommendations.

[0126] For each set of input information, the language model performs semantic matching according to the semantic matching process indicated by the plug-in matching strategy, which may include: performing semantic understanding on the plug-in requirement information to obtain the requirement semantic features corresponding to the plug-in requirement information; performing semantic understanding on the plug-in document of each target plug-in to obtain the document semantic features corresponding to the plug-in document of each target plug-in; performing feature matching on the requirement semantic features and the document semantic features corresponding to the plug-in document of each target plug-in to predict the recommendation probability of each target plug-in to the business object.

[0127] ② Based on the recommendation probability of at least one target plug-in, the process of selecting the first recommended plug-in from at least one target plug-in may include: comparing the recommendation probability with a second probability threshold (the second probability threshold may be set based on an empirical value), and determining the target plug-in whose corresponding recommendation probability is greater than or equal to the second probability threshold as the first recommended plug-in; or, sorting the at least one target plug-in in descending order of recommendation probability, and determining the target plug-in ranked in the top M positions as the first recommended plug-in, where M is a positive integer.

[0128] In summary, the above steps s11-s12 introduce the method of first selecting a portion of the target plug-in documents that are similar to the plug-in requirement information from the plug-in documents of multiple candidate plug-ins, and then performing semantic matching. Optionally, semantic matching can also be performed directly between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins. This process may include: the plug-in document of each candidate plug-in can be spliced ​​with the plug-in requirement information to obtain a set of input information, resulting in multiple sets of input information in total; the multiple sets of input information can be input into a language model, and the language model can process the multiple sets of input information in parallel, and finally output the recommendation probabilities corresponding to the multiple sets of input information, that is, the recommendation probabilities of the multiple candidate plug-ins.

[0129] For each set of input information, the language model performs semantic matching according to the semantic matching process indicated by the plug-in matching strategy, which may include: semantic understanding of the plug-in requirement information to obtain the requirement semantic features corresponding to the requirement information; semantic understanding of the plug-in document of each candidate plug-in to obtain the document semantic features corresponding to the plug-in document of each candidate plug-in; feature matching of the requirement semantic features with the document semantic features corresponding to the plug-in document of each candidate plug-in to predict the recommendation probability of each candidate plug-in to the business object; and selection of a first recommended plug-in from multiple candidate plug-ins based on the recommendation probabilities of multiple candidate plug-ins. The process of selecting a first recommended plug-in from multiple candidate plug-ins based on the recommendation probabilities of multiple candidate plug-ins is similar to the process of selecting a first recommended plug-in from at least one target plug-in based on the recommendation probability of at least one target plug-in, and will not be repeated here.

[0130] S304: Determine, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information.

[0131] S305: Perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

[0132] In step S305, the plug-in recommendation processing process may include: sending the plug-in information (for example, the plug-in name) and the first recommended content of the first recommended plug-in to the plug-in market client of the business object, so that the plug-in market client of the business object outputs the plug-in information and the first recommended content of the first recommended plug-in to the business object.

[0133] In the embodiment of the present application, a search enhancement generation technology is introduced in the plug-in market for plug-in recommendation based on the needs of business objects. The advantages of this are: through an intelligent search mechanism based on needs, the retrieval model is used to quickly retrieve the plug-in documents that best match the needs from the plug-in database, and the language model is used to process and recommend this information. Compared with the traditional manual screening method, this method can greatly shorten the time for business objects to find suitable plug-ins and improve the efficiency of plug-in selection; and, through the search enhancement generation technology, combined with the needs and the relevant plug-in documents in the plug-in database library, more accurate plug-in recommendations can be provided. The collaborative work of the language model and the retrieval model makes the plug-in recommendation not only based on keywords, but also able to deeply understand the specific needs and recommend plug-ins that are more in line with the actual usage scenario, avoiding the problem of mismatch between plug-in functions and needs in traditional methods. In addition, the embodiment of the present application adopts an innovative method of collaborative work of the language model and the retrieval model, and provides personalized recommendations by combining the needs in the plug-in market with the plug-in information. The language model is used as a fixed module, and the retrieval model is responsible for retrieving relevant plug-in information from the plug-in database based on needs and providing this information as context to the language model. Through this simplified design, plug-in recommendations can be accurately matched according to the specific needs of business objects without the need for complex adjustments or fine-tuning of the language model, significantly improving the efficiency and accuracy of recommendations.

[0134] The present application provides an information processing method, which includes the following contents: optimization of recommendation probability by similarity, participation of target software of the plug-in to be installed in the plug-in recommendation process, determination process of the first recommended content, and adjustment and optimization of the plug-in matching strategy. The information processing method can be executed by a computer device, which can be, for example, the server 102 in the information processing system. Figure 4 As shown, the information processing method may include but is not limited to the following steps S401 to S408:

[0135] S401: Obtain plug-in requirement information of a business object, and obtain plug-in documents of multiple candidate plug-ins.

[0136] In the embodiment of the present application, the execution process of step S401 is the same as the above Figure 3 The execution process of step S301 in the embodiment shown is the same, and the details can be found in the above Figure 3 The description of step S301 in the illustrated embodiment will not be repeated here.

[0137] S402: Obtain a plug-in matching strategy, where the plug-in matching strategy is used to indicate a semantic matching process between plug-in requirement information and plug-in documents.

[0138] In the embodiment of the present application, the execution process of step S402 is the same as the above Figure 3The execution process of step S302 in the embodiment shown is the same, and the details can be found in the above Figure 3 The relevant description of step S302 in the illustrated embodiment will not be repeated here.

[0139] S403 , calling a retrieval model to filter out at least one target plug-in document from the plug-in documents of the multiple candidate plug-ins based on similarities between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins.

[0140] In the embodiment of the present application, the execution process of step S403 is the same as the above Figure 3 The execution process of sub-step s11 in the embodiment shown is the same, and the details can be found in the above Figure 3 The relevant description of sub-step s11 in the illustrated embodiment will not be repeated here.

[0141] Optionally, in addition to inputting plug-in requirement information through the plug-in requirement input operation, the business object can also select the target software of the plug-in to be installed through the software selection operation. In this case, the software type of the target software can also be considered during the similarity screening process in step S403. A first screening can be performed based on the software type of the target software of the plug-in to be installed, and a second screening can be performed based on the similarity to screen out the plug-in document of at least one target plug-in. In this way, the plug-in document of the screened at least one target plug-in can not only be similar to the requirement, but also match the software type. Specifically, the software type of the target software of the plug-in to be installed can be obtained; the software type of the target software can be matched with the plug-in service types of multiple candidate plug-ins to screen out multiple reference plug-in documents from the plug-in documents of the multiple candidate plug-ins; and based on the similarity between the plug-in requirement information and the plug-in documents of the multiple reference plug-ins, the plug-in document of at least one target plug-in can be screened out from the plug-in documents of the multiple reference plug-ins.

[0142] In detail, the software type of the target software refers to the functional type of the target software. For example, the software type may include video type, news type or game type, etc. The plug-in service type of the candidate plug-in refers to the software type served by the candidate plug-in type. For example, the candidate plug-in is used to provide services for video type software, for news type software, or for game type software, etc. Type matching processing refers to comparing whether the software type of the target software is the same as the plug-in service type of the candidate plug-in, and the plug-in service type of the reference plug-in obtained by screening is the same as the software type of the target software. It is easy to think that by screening the reference plug-ins whose corresponding plug-in service type is the same as the software type of the target software, the plug-in recommended to the business object can provide more targeted functional extensions for the target software, which is conducive to improving the plug-in recommendation effect.

[0143] The process of selecting at least one target plug-in document from the plug-in documents of multiple reference plug-ins based on the similarity between the plug-in requirement information and the plug-in documents of multiple reference plug-ins is similar to the process of selecting at least one target plug-in document from the plug-in documents of multiple candidate plug-ins based on the similarity between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and will not be repeated here.

[0144] S404 , calling the language model to perform semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and predicting the recommendation probability of recommending at least one target plug-in to the business object.

[0145] In the embodiment of the present application, the execution process of step S404 is the same as the above Figure 3 The process of predicting the recommendation probability based on semantic matching in sub-step s12 of the embodiment shown is the same. For details, please refer to the above Figure 3 The process of predicting the recommendation probability based on semantic matching in sub-step s12 of the illustrated embodiment will not be described in detail here.

[0146] In the description of step S404 above, semantic matching can be performed between the plug-in requirement information and the entire content of the target plug-in's plug-in document. Optionally, semantic matching can be performed between the plug-in requirement information and similar document content within the target plug-in's plug-in document that is similar to the plug-in requirement information. Specifically, similar document content similar to the plug-in requirement information can be determined within the plug-in document of each target plug-in. Following the semantic matching process indicated by the plug-in matching policy, semantic matching can be performed between the plug-in requirement information and similar document content corresponding to at least one target plug-in, and the probability of recommending each target plug-in to the business object can be predicted.

[0147] Among them, the process of determining similar document content similar to the plug-in requirement information in the plug-in document of each target plug-in may include: the plug-in document of the target plug-in may be subjected to document content division processing to obtain multiple document contents, where the content division processing may be performed according to content division units (for example, paragraphs, sentences, etc. as content division units), for example, a paragraph is regarded as a document content, or a sentence is regarded as a document content, etc.; the similarity between the plug-in requirement information and each document content may be calculated; the similarity calculation method here is similar to the similarity calculation method between the plug-in requirement information and the plug-in document of the candidate plug-in, both of which are vector similarities and will not be repeated here; similar document content similar to the plug-in requirement information may be selected from each document content based on the similarity between the plug-in requirement information and each document content, for example, document content with a similarity greater than a similarity threshold may be selected as similar document content, or, for example, Q document contents with the highest similarity may be selected as similar document content, where Q is a positive integer.

[0148] The process of predicting the recommendation probability by semantic matching between the plug-in requirement information and the similar document content corresponding to at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy is similar to the process of predicting the recommendation probability by semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and will not be repeated here. It is easy to think that similar document content similar to the plug-in requirement information is selected from the plug-in document of at least one target plug-in. This part of the content is what needs to be focused on in semantic matching. Performing semantic matching between the plug-in requirement information and the similar document content of at least one target plug-in, that is, performing semantic matching between the plug-in requirement information and part of the relevant document content in the plug-in document of at least one target plug-in, rather than performing semantic matching between the plug-in requirement information and all the document content in the plug-in document of at least one target plug-in, can improve the efficiency of semantic matching, thereby helping to improve the efficiency of plug-in recommendation.

[0149] Optionally, in addition to inputting plug-in requirement information through the plug-in requirement input operation, the business object may also select the target software for the plug-in to be installed through the software selection operation. In this case, the software description information of the target software may also be considered during the semantic matching process in step S404. Specifically, the plug-in requirement information may be adjusted based on the software description information of the target software to obtain adjusted plug-in requirement information; and semantic matching may be performed between the adjusted plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching policy, thereby selecting a first recommended plug-in from the at least one target plug-in.

[0150] In detail, software description information refers to information used to describe the software attributes of the target software. The software description information here may include but is not limited to at least one of the following: software type, software function, software development information (for example, software development environment, development language, etc.), software version information, and software operating environment information. Demand adjustment processing refers to semantic enhancement processing of plug-in requirement information through the software description information of the target software, and integrating the semantics of the software description information of the target software into the plug-in requirement information. For example, the plug-in requirement information is "I want to do advertisement recognition", and the software description information of the target software is "The target software is a news software..." After demand adjustment processing (i.e., semantic enhancement processing), the adjusted plug-in requirement information is "The target software is a news software... I want to perform advertisement recognition in the target software." It is not difficult to see that by considering the software description information of the target software in the semantic matching process, the recommended plug-in can be better adapted to the software attributes of the target software, thereby improving the accuracy of plug-in recommendation.

[0151] Optionally, the recommendation probability predicted by the language model can not only be used to generate recommendation results, but also can be used as supervisory information to fine-tune the retrieval model. Specifically, the fine-tuning information of the retrieval model can be determined based on the difference between the recommendation probability of at least one target plug-in and the selection probability of at least one target plug-in. The fine-tuning information here can be, for example, the KL (Kullback-Leibler Divergence) divergence between the recommendation probability of at least one target plug-in and the selection probability of at least one target plug-in; the retrieval model can be fine-tuned based on the fine-tuning information of the retrieval model. That is to say, in the actual plug-in recommendation scenario, the embodiment of the present application can use the recommendation probability predicted by the language model as supervisory information to continuously fine-tune the retrieval model and continuously optimize the retrieval model, so that the retrieval model can be self-optimized without changing the model structure, which is conducive to improving the accuracy of similarity calculation, thereby improving the accuracy of plug-in document screening based on similarity.

[0152] S405 , calling a language model to optimize the recommendation probability of at least one target plug-in according to the similarity between the plug-in requirement information and the plug-in document of at least one target plug-in, to obtain an optimized probability of the at least one target plug-in.

[0153] In step S405, in order to ensure that different plug-in documents have an uneven impact on the final recommendation results, a weighted fusion strategy is introduced after obtaining the recommendation probability of at least one plug-in. The so-called weighted fusion strategy refers to assigning a weight to each target plug-in based on the similarity between the plug-in document of each target plug-in and the plug-in requirement information, and optimizing the recommendation probability of each target plug-in based on the assigned weight to obtain the optimized probability of each target plug-in.

[0154] Specifically, the optimization process may include: performing weight assignment processing on at least one target plug-in based on the similarity between the plug-in requirement information and the plug-in document of at least one target plug-in to obtain the optimization weight of at least one target plug-in; optimizing the recommendation probability of at least one target plug-in based on the optimization weight of at least one target plug-in to obtain the optimization probability of at least one target plug-in.

[0155] During the above optimization process:

[0156] ① The optimization weight is proportional to the similarity. The weight distribution process can be seen in the following formula 3:

[0157]

[0158] In the above formula 3, D′={d1,d2,…,d t} represents a second plug-in document set consisting of plug-in documents of at least one target plug-in; λ(d j |x) represents the optimization weight of any target plug-in (jth target plug-in) in the second plug-in document set D′, where j is a positive integer less than or equal to t; s(d j |x) represents the similarity between the plug-in requirement information and the j-th target plug-in.

[0159] ② The optimization process based on optimization weights may include: weighting the recommendation probability of each target plug-in according to the optimization weight of each target plug-in to obtain the weighted probability of each target plug-in; fusing the weighted probabilities of at least one target plug-in to obtain a weighted fusion probability, where the fusion process refers to adding the weighted probabilities of at least one target plug-in; and determining the proportion of the weighted probability of each target plug-in in the weighted fusion probability as the optimization probability of each target plug-in. The above optimization process can be specifically referred to in the following formulas 4 and 5:

[0160]

[0161] In the above formulas 4 and 5, D′={d1, d2, …, d t} represents a second plug-in document set consisting of plug-in documents of at least one target plug-in, d j represents the jth target plug-in, which is any target plug-in in the second plug-in document set D′, where j is a positive integer less than or equal to t; λ(d j |x) represents the optimization weight of the jth target plug-in; P(y|x,d j ) represents the recommendation probability of the jth target plug-in, P(y|x,d j )·λ(d j |x) represents the weighted probability of the jth target plug-in; P(y|x,D ′ ) represents the weighted fusion probability; P(y ′ |x,d j ) represents the optimization probability of the jth target plug-in.

[0162] In step S405, the recommendation probability of at least one target plug-in is optimized by the similarity of at least one target plug-in, which is equivalent to fusing the similarity and the recommendation probability. In this way, the optimization probability of the target plug-in can more accurately reflect the possibility of recommending the target plug-in to the business object, thereby improving the accuracy of plug-in recommendation.

[0163] S406 : Select a first recommended plug-in from the at least one target plug-in according to the optimization probability of the at least one target plug-in.

[0164] In step S406, the first recommended plug-in selection process based on the optimization probability may include: comparing the optimization probability with a third probability threshold, and determining the target plug-in whose corresponding optimization probability is greater than or equal to the third probability threshold (the third probability threshold may be set based on an empirical value) as the first recommended plug-in; or, sorting at least one target plug-in in descending order of optimization probability, and determining the target plug-in ranked in the top M positions as the first recommended plug-in, where M is a positive integer.

[0165] S407: Determine, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information.

[0166] In step S407, the process of determining the first recommended content may include: obtaining the recommended content type corresponding to the plug-in requirement information; performing content screening processing on the plug-in document of the first recommended plug-in according to the recommended content type to obtain the filtered document content; performing content matching processing between the plug-in requirement information and the filtered document content to determine the first recommended content semantically related to the plug-in requirement information from the filtered document content.

[0167] In the process of determining the first recommended content:

[0168] ① The recommended content type can be preset. Preset means that the recommended content type is the content type specified in the plug-in market client. For example, if the plug-in market client specifies two types of content: output plug-in function and plug-in effect display, then the recommended content type can include plug-in function content type and plug-in effect display content type. Alternatively, the recommended content type can be obtained by content recognition of the plug-in requirement information. Here, content recognition can refer to extracting content type keywords from the plug-in requirement information, and the content type indicated by the content type keyword can be determined as the recommended content type. For example, if the plug-in requirement information is "a plug-in with advertising recognition function is required", the content type keyword "function" can be extracted, and the content type indicated by the content type keyword is the plug-in function content type.

[0169] ② The process of content screening processing may include: performing document content division processing on the plug-in document of the first recommended plug-in to obtain multiple document contents. The content division processing here may be divided according to content division units (for example, paragraphs, sentences, etc. as content division units), for example, taking a paragraph as a document content, or taking a sentence as a document content, etc.; performing content type identification on each document content to obtain the content type to which each document content belongs; and determining document content whose content type is the recommended content type as the screened document content. For example, if the recommended content type is the plug-in function content type, the document content whose content type is the plug-in function content type among the multiple document contents may be determined as the screened document content.

[0170] ③ The process of content matching processing may include: predicting the semantic matching degree between the plug-in requirement information and each filtered document content based on the semantic correlation between the plug-in requirement information and each filtered document content, and determining the filtered document content with the highest semantic matching degree as the first recommended content semantically related to the plug-in requirement information.

[0171] It is not difficult to see that in the process of determining the first recommended content, on the one hand, it is considered that the content type of the first recommended content is the same as the recommended content type corresponding to the plug-in requirement information, and the semantics of the first recommended content matches the semantics of the plug-in requirement information. In this way, the first recommended content of the first recommended plug-in can be consistent with the needs of the business object, and highly matched, which can improve the accuracy of plug-in recommendations.

[0172] S408: Perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

[0173] In step S408, the plug-in recommendation processing process may include: sending the plug-in information (for example, the plug-in name) and the first recommended content of the first recommended plug-in to the plug-in market client of the business object, so that the plug-in market client of the business object outputs the plug-in information and the first recommended content of the first recommended plug-in to the business object.

[0174] More specifically, the plug-in market client of the business object can output the plug-in information and first recommended content of the first recommended plug-in to the business object according to the recommended content type. For example, if the plug-in market client specifies the recommended content types as plug-in function content type and plug-in effect content type, the first recommended content under the plug-in function content type can be output in the display area corresponding to the plug-in function content type in the plug-in market client of the business object, and the first recommended content under the plug-in effect content type can be output in the display area corresponding to the plug-in effect content type in the plug-in market client of the business object.

[0175] In an embodiment of the present application, by combining the intelligent retrieval of the retrieval model with the semantic matching generation of the language model, it is possible to effectively filter out the plug-ins most relevant to the needs of the business object from a large number of plug-ins, reducing the information overload problem faced by the business object. The business object can quickly make decisions based on the plug-in recommendation results, without having to spend a lot of time screening irrelevant plug-ins. In addition, the embodiment of the present application can also realize personalized plug-in recommendations. Based on the retrieval enhancement generation technology, combined with the personalized understanding of the language model and the retrieval ability of the retrieval model, it can provide personalized plug-in recommendations based on the needs of the business object. Compared with the traditional recommendation method based on historical plug-in preferences or content, the embodiment of the present application can better adapt to the diverse plug-in acquisition needs of different business objects and provide more accurate recommendation results. In addition, the embodiment of the present application introduces a context-based plug-in recommendation mechanism into the traditional RAG method. The traditional RAG method usually inputs the plug-in requirement information and the retrieved plug-in document into the generation module, and then generates the recommendation results. However, simply splicing the retrieved plug-in document and the requirements together and generating them often cannot fully consider the different impacts of multiple documents. Therefore, the present invention adopts a multi-document parallel processing and weighted fusion strategy to make full use of the information of different retrieval results, thereby improving the language model's prediction accuracy for recommendation probability. In general, the embodiment of the present application can effectively solve the problems of inefficiency, inaccuracy, information overload and lack of personalized recommendations in the existing plug-in selection method through intelligent retrieval enhancement generation technology, greatly improving the efficiency and accuracy of plug-in selection in the plug-in market, and significantly improving the plug-in recommendation experience.

[0176] above Figure 3 and Figure 4 The illustrated embodiment introduces the technical process of plug-in recommendation. The following describes the plug-in recommendation scenario when a business object uses the plug-in market.

[0177] Scenario 1: The business object enters plug-in requirement information.

[0178] In scenario one, the business object can input plug-in requirement information through the plug-in requirement input operation. The plug-in market can recommend the first recommended plug-in that matches the plug-in requirement information and the first recommended plug-in that is semantically related to the plug-in requirement information to the business object through the plug-in recommendation technical process.

[0179] For example, Figure 5aAs shown, the plug-in market client of the business object may include a query input box 501, a search control 502 and a plug-in type selection label list 503, and the plug-in type selection label list 503 may include selection labels of multiple plug-in types; if the business object enters a plug-in requirement description text in the query input box 501 and triggers the search control 502, the input plug-in requirement description text may be determined as plug-in requirement information; if the business object selects a selection label of a target plug-in type in the plug-in type selection label list 503, the selected target plug-in type may be determined as the plug-in requirement information. Figure 5a For example, the business object enters a plug-in requirement description text in the query input box 501. The entered plug-in requirement description text is "I have some content that I want to do advertisement identification."

[0180] Taking the case where the number of the first recommended plug-in is one and the recommended content type is specified as the plug-in function content type and the plug-in effect content type in the plug-in market client as an example, in response to the triggering operation of the search control 502, the plug-in information and the first recommended content of the first recommended plug-in can be output in the plug-in market client of the business object. For example, the plug-in information 5041 of the first recommended plug-in and the first recommended content 5042 under the plug-in function content type can be output in the content area 504 corresponding to the plug-in function content type in the plug-in market client of the business object, and the first recommended content 5051 under the plug-in effect content type can be output in the content area 505 corresponding to the plug-in effect content type in the plug-in market client of the business object. In addition, Figure 5a As shown, the plug-in market client of the business object can also output the thinking process 5061 of the language model in the content area 506 corresponding to the thinking process.

[0181] It can be seen that scenario one can accurately recommend plug-ins to business objects according to their plug-in acquisition requirements, and can also display rich information related to the first recommended plug-in to the business objects, thereby improving the plug-in recommendation effect.

[0182] Scenario 2: The business object enters the plug-in requirement information and selects the target software.

[0183] In scenario two, the business object can input plug-in requirement information through the plug-in requirement input operation, and can select the target software for the plug-in to be installed through the software selection operation. The plug-in market can recommend to the business object the first recommended plug-in that matches the plug-in requirement information and is targeted at the target software, as well as the first recommended plug-in that is semantically related to the plug-in requirement information through the plug-in recommendation technical process.

[0184] For example, Figure 5bAs shown, the plug-in market client of the business object may further include a software list 507, and the software list 507 may include software identifiers of multiple software that support the installation of plug-ins (for example, the software identifier is Figure 5b In response to the software selection operation on the software list 507, the software selection operation selects the software identifier 508 of the target software, and determines that the target software is selected as the target software to install the plug-in, that is, the target software is the software for which the plug-in is to be installed; and, in response to the plug-in requirement description text "I have some content that I want to do advertising identification" input by the business object in the query input box 501, the plug-in requirement description text is determined as the plug-in requirement information. In response to the triggering operation on the search control 502, the plug-in information and the first recommended content of the first recommended plug-in can be output in the plug-in market client of the business object, and the output plug-in information and the first recommended content of the first recommended plug-in are consistent with the plug-in requirement information of the business object. Figure 5a Similar, no further description is given here.

[0185] It can be seen that scenario two can accurately recommend plug-ins to business objects based on their plug-in acquisition and installation requirements, and based on the software of the plug-in to be installed selected by the business objects. In addition, it can also display rich information related to the first recommended plug-in to the business objects, thereby improving the plug-in recommendation effect.

[0186] Scenario 3: Plug-in recommendation conversation system with business objects.

[0187] In scenario three, plug-in recommendations are implemented based on the conversation system. Specifically, a business object can input plug-in requirement information and the software identifier of the target software to be installed through the business object's conversation message. The conversation system then outputs the first recommended plug-in and first recommended content to the business object through the conversation system's response message.

[0188] For example, Figure 5c As shown, the business object inputs the plug-in requirement information and the software identifier of the target software of the plug-in to be installed in the question-answering system of the plug-in market client through the first session information 509, for example, Figure 5c "I want to identify advertisements in XX news software"; the conversation system of the plug-in market client can output a first response message 510 to the business object, and the first response message 510 may include the plug-in information 5041 of the first recommended plug-in, the first recommended content 5042 and the thinking process 5061 of the language model.

[0189] Optionally, in scenario three, recommendation feedback information of the business object can be obtained, and the semantic matching process indicated by the plug-in matching strategy can be updated according to the recommendation feedback information to obtain an updated plug-in matching strategy; semantic matching can be performed between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins according to the semantic matching process indicated by the updated plug-in matching strategy, and a second recommended plug-in can be selected from the multiple candidate plug-ins; second-first recommendation content semantically related to the plug-in requirement information can be determined in the plug-in document of the second-first recommended plug-in; and plug-in recommendation processing can be performed on the business object based on the second recommended plug-in and the second recommended content.

[0190] The update process may include at least one of the following: updating the steps in the semantic matching process indicated by the plug-in matching strategy according to the recommendation feedback information, updating the execution order between steps, adding steps, reducing steps, etc.

[0191] For example, Figure 5c As shown, the business object inputs the recommendation feedback information in the session system of the plug-in market client through the second session information 511, for example, Figure 5c "My plug-in acquisition preferences were not considered" in the plug-in market client. Based on this, the step "considering the semantics expressed in the plug-in document of the historically acquired plug-in and extracting the plug-in preference features from it" can be added to the semantic matching process indicated by the plug-in matching strategy according to the recommendation feedback information, and the step "considering the feature matching degree between the demand semantic features, plug-in preference features, and document semantic features" can be updated. Plug-in recommendations can then be re-performed according to the updated plug-in matching strategy. The business object outputs the re-recommendation result 512, i.e., the plug-in information of the second recommended plug-in and the second recommended content, in the session system of the plug-in market client through the second response message 512.

[0192] It should be noted that Figure 5c Taking one round of conversation (one conversation message + one response message constitutes one round of conversation) as an example, the collection of recommendation feedback information and plug-in re-recommendation are introduced. In actual usage scenarios, multiple rounds of conversations can be conducted, that is, recommendation feedback can be conducted multiple times and plug-in re-recommendation can be conducted multiple times. The embodiments of this application do not limit this.

[0193] It can be seen that scenario three can make use of the conversation system to recommend plug-ins, which increases the fun of plug-in recommendations. In addition, the introduction of the conversation system can realize recommendation feedback and plug-in re-recommendation, continuously optimize the recommendation results, and further improve the accuracy of plug-in recommendations.

[0194] The embodiment of the present application provides an information processing method, which includes the training process of an information processing model (including a retrieval model and a language model). The information processing method can be executed by a computer device, which can be, for example, a server 102 in an information processing system. Figure 6 As shown, the information processing method may include but is not limited to the following steps S601 to S605:

[0195] S601: Acquire sample data for training an information processing model, where the sample data includes sample requirement information and plug-in documents of a plurality of sample plug-ins.

[0196] In step S601, the sample requirement information may be any plug-in requirement information in a historical plug-in recommendation scenario. The plug-in documents of the multiple sample plug-ins may be the plug-in documents of all candidate plug-ins in the plug-in database, or the plug-in documents of the multiple sample plug-ins may be selected from the plug-in documents of multiple candidate plug-ins in a plug-in recommendation process associated with the sample requirement information based on the similarity between the sample requirement information and the plug-in documents of the multiple candidate plug-ins. This embodiment of the present application is not limited to this.

[0197] S602 : Calling a retrieval model to perform similarity matching between the sample requirement information and plug-in documents of multiple sample plug-ins, and obtaining selection probabilities of the retrieval model for the plug-in documents of the multiple sample plug-ins.

[0198] In step S602, the retrieval model obtains the selection probability through similarity matching, which may include: performing similarity matching between the sample requirement information and the plug-in documents of multiple sample plug-ins to obtain the similarity between the sample requirement information and the plug-in documents of multiple sample plug-ins, and determining the similarity between the sample requirement information and the plug-in documents of multiple sample plug-ins as the selection probability of the retrieval model for the plug-in documents of multiple sample plug-ins. The similarity calculation process here can be referred to the above Figure 3 The similarity calculation process in sub-step s11 of the illustrated embodiment will not be described in detail here.

[0199] Alternatively, the retrieval model may determine the selection probability through similarity matching, which may include: performing similarity matching between the sample requirement information and the plug-in documents of multiple sample plug-ins to obtain the similarity between the sample requirement information and the plug-in documents of multiple sample plug-ins; obtaining a similarity adjustment factor, and performing distribution adjustment processing on the similarity between the sample requirement information and the plug-in documents of multiple sample plug-ins according to the distribution adjustment direction corresponding to the similarity adjustment factor to obtain the selection probability of the retrieval model for the plug-in documents of multiple sample plug-ins. The similarity calculation process here can be found in the above Figure 3The similarity calculation process in sub-step s11 of the embodiment shown is not described here in detail; the distribution adjustment process here can be found in the above Figure 3 The distribution adjustment process in sub-step s11 of the illustrated embodiment will not be described in detail here.

[0200] S603 , calling the language model to perform semantic matching between the sample requirement information and the plug-in documents of the multiple sample plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and predicting the recommendation probability of the multiple sample plug-ins.

[0201] In step S603, the language model performs semantic matching based on the semantic matching process indicated by the plug-in matching strategy to predict the recommendation probability. This process may include: concatenating the sample requirement information with the plug-in document of each sample plug-in to obtain sample input information corresponding to each sample plug-in, that is, obtaining multiple sets of sample input information. The language model may process the multiple sets of sample input information in parallel to obtain the recommendation probability of the sample plug-in corresponding to each set of sample input information.

[0202] The language model's processing of any set of sample input information may include: the language model may perform semantic understanding on the sample requirement information to obtain sample requirement semantic features corresponding to the sample requirement information, perform semantic understanding on the plug-in document of the sample plug-in to obtain sample document semantic features of the plug-in document of the sample plug-in, perform feature matching on the sample requirement semantic features and the sample document semantic features of the plug-in document of the sample plug-in, and predict the recommendation probability of the sample plug-in.

[0203] S604 : Determine loss information of the information processing model according to the difference between the selection probabilities of the plug-in documents of the multiple sample plug-ins and the recommendation probabilities of the multiple sample plug-ins.

[0204] In step S604, loss information of the information processing model can be determined based on the difference between the selection probabilities of the plug-in documents of the multiple sample plug-ins and the recommendation probabilities of the multiple sample plug-ins. Here, the loss information of the information processing model can be the KL divergence between the selection probabilities of the plug-in documents of the multiple sample plug-ins and the recommendation probabilities of the multiple sample plug-ins. In detail, the calculation method of the loss information can be referred to the following formula 6:

[0205]

[0206] In the above formula 6, B represents the sample plug-in set, which is a set of plug-in documents of each sample plug-in in the sample data; L represents loss information; KL represents KL divergence calculation; x represents sample requirement information, d represents any sample plug-in in the sample plug-in set; P R (d, x) represents the probability of the retrieval model selecting the plug-in document of any sample plug-in; PLM (d, x) represents the recommendation probability of any sample plug-in. In other words, the difference between the selection probability of the plug-in documents of multiple sample plug-ins and the recommendation probability of multiple sample plug-ins can be measured using KL divergence.

[0207] S605: Train the retrieval model in the information processing model according to the loss information.

[0208] In step S605, training the retrieval model in the information processing model based on the loss information means optimizing the model parameters of the retrieval model in the direction of reducing the loss information. The direction of reducing the loss information mentioned here refers to the model optimization direction with the goal of minimizing the loss information; by optimizing the model in this direction, the loss information generated by the retrieval model after each optimization must be less than the loss information generated by the retrieval model before optimization. For example, if the loss information calculated this time is 0.85, then after optimizing the retrieval model in the direction of reducing the loss information, the loss value generated by optimizing the retrieval model should be less than 0.85.

[0209] In other words, the present embodiment uses KL divergence to measure the difference between the selection probability of plugin documents for multiple sample plugins and the recommendation probability of multiple sample plugins. The optimization goal of the retrieval model is to minimize the difference between the selection probability of plugin documents for multiple sample plugins and the recommendation probability of multiple sample plugins, that is, to minimize the difference between the output of the retrieval model and the language model. The benefit of this training is that the plugin documents selected by the retrieval model are documents that are beneficial to the language model, thereby ensuring that the recommended plugins are more in line with the needs of the business object.

[0210] In an embodiment of the present application, by using the output of the language model as a supervisory signal to train the retrieval model, the difference between the selection probability output by the retrieval model and the recommendation probability output by the language model can be narrowed, so that the plug-in documents screened by the retrieval model based on the selection probability can be highly matched with the requirements, thereby improving the retrieval relevance and accuracy of the retrieval model. The retrieval model is optimized through the scoring function of the language model to better understand the needs of the business object and retrieve the most relevant plug-ins from the plug-in database, ensuring that the retrieved plug-in information is most effective in solving the needs. Through this optimization mechanism, plug-ins can be more intelligently and accurately recommended based on the needs of the business object, improving the relevance and accuracy of plug-in selection.

[0211] In summary Figure 3-Figure 6The content of the embodiment shown in the figure, the embodiment of the present application introduces search enhancement generation technology in the plug-in market, which can intelligently recommend plug-ins that are adapted to the needs of the users (i.e., business objects) of the plug-in market; by combining the collaborative work of the language model and the retrieval model, it effectively solves the problems of inefficiency, information overload, and inaccurate matching between plug-in functions and business object needs in the traditional plug-in market. Figure 7 The information processing method provided in the embodiments of this application is summarized as follows:

[0212] First, the optimization of the retrieval model is specifically reflected in the following two points:

[0213] (1) The retrieval model can be used to perform document retrieval. Document retrieval refers to screening plug-in documents that match the plug-in requirement information of the business object, corresponding to the above Figure 3-Figure 6 In the illustrated embodiment, in a plug-in recommendation scenario, the retrieval model may be used to filter at least one target plug-in document having similar plug-in requirement information from a plurality of candidate plug-in documents included in a plug-in database.

[0214] (2) The retrieval model can calculate the selection probability. The retrieval model can calculate the similarity between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and introduce a similarity adjustment factor to adjust the similarity between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins to obtain the selection probability of the plug-in documents of multiple candidate plug-ins. The selection probability of the plug-in documents of multiple candidate plug-ins can be used for document retrieval.

[0215] Second, the language model assists the retrieval model in training, which is specifically reflected in the following two points:

[0216] (1) Divergence optimization. The output of the language model is used as supervisory information to train the retrieval model. The KL divergence is used to measure the difference between the output of the language model and the output of the retrieval model. The training goal is to reduce the difference between the output of the language model and the output of the retrieval model.

[0217] (2) Generation and optimization of language models. For details, please refer to point 3 below.

[0218] Point 3: Context-based plugin recommendations are specifically reflected in the following two points:

[0219] (1) Parallel processing of multiple documents. The context here refers to the plugin document of at least the target plugin selected by the retrieval model from the plugin documents of multiple candidate plugins. The language model can perform semantic matching between the plugin requirement information and the plugin document of each target plugin in parallel, and predict the probability of recommending each target plugin to the business object.

[0220] (2) Weighted fusion. Weighted fusion refers to using the similarity between the plug-in requirement information and the plug-in document of at least one target plug-in to assign an optimization weight to at least one target plug-in. Based on the optimization weight of at least one target plug-in, the recommendation probability of at least one target plug-in output by the language model is optimized to obtain the optimization probability of at least one target plug-in.

[0221] Point 4: Algorithm selection and advantages. The advantages of the algorithm are specifically reflected in the following four points:

[0222] (1) Efficiency. The retrieval model and the language model work independently, avoiding fine-tuning the language model and saving computing resources. In the entire model, only the retrieval model is trained, which improves the efficiency of model training. In addition, by adopting retrieval enhancement generation technology, the first recommended plug-in and the first recommended content that are highly relevant to the needs can be quickly retrieved by simply inputting the requirements, avoiding the tedious process of manual screening in traditional methods. Business objects do not need to browse and evaluate a large number of plug-ins, but can quickly obtain the most relevant plug-in recommendations, which greatly improves the efficiency of plug-in selection.

[0223] (2) Accuracy. The language model can more comprehensively consider multiple retrieved plug-in documents through multi-document parallel processing and weighted fusion strategies, thereby improving the accuracy of plug-in recommendations. In addition, by introducing auxiliary training of the language model and optimizing the retrieval model using the language model, it ensures that the retrieved plug-in information is highly matched with the specific needs of the business object. By minimizing the KL divergence between the selection probability output by the retrieval model and the recommendation probability generated by the language model, plug-ins that meet the needs of the business object can be accurately recommended, avoiding the problem of inaccurate matching between plug-in functions and needs in traditional recommendation methods.

[0224] (3) Personalization. The weighted fusion mechanism can dynamically adjust the recommendation results based on the matching degree between each plug-in document and the needs of the business object, thus achieving personalized recommendations.

[0225] (4) Flexibility. Since there is no need to fine-tune the language model, the information processing method can be easily applied to different plug-in markets and has strong adaptability.

[0226] The above describes in detail the method of the embodiment of the present application. In order to facilitate better implementation of the above scheme of the embodiment of the present application, the device of the embodiment of the present application is provided below accordingly.

[0227] See Figure 8 , Figure 8 This is a structural diagram of an information processing device provided in an embodiment of the present application. The information processing device can be set in the computer device provided in an embodiment of the present application, and the computer device can be a server. Figure 8The information processing device shown may be a computer program running on a computer device, and the information processing device may be used to execute Figure 3 、 Figure 4 or Figure 6 Some or all of the steps in the method embodiment shown. Figure 8 , the information processing device may include the following units:

[0228] An acquiring unit 801 is configured to acquire plug-in requirement information of a business object and acquire plug-in documents of multiple candidate plug-ins;

[0229] The acquisition unit 801 is further configured to acquire a plug-in matching strategy, where the plug-in matching strategy indicates a semantic matching process between the plug-in requirement information and the plug-in document.

[0230] Processing unit 802 is configured to perform semantic matching between the plug-in requirement information and plug-in documents of multiple candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and select a first recommended plug-in from the multiple candidate plug-ins;

[0231] The processing unit 802 is further configured to determine, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information;

[0232] The processing unit 802 is further configured to perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

[0233] In one implementation, processing unit 802 is configured to perform semantic matching between plug-in requirement information and plug-in documents of multiple candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and to select a first recommended plug-in from the multiple candidate plug-ins by specifically executing the following steps:

[0234] Based on the similarity between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins, screening out at least one plug-in document of the target plug-in from the plug-in documents of the plurality of candidate plug-ins;

[0235] According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the plug-in requirement information and the plug-in document of at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

[0236] In one implementation, processing unit 802 is configured to perform semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching policy, and to select a first recommended plug-in from the at least one target plug-in by performing the following steps:

[0237] Performing semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and predicting a probability of recommending at least one target plug-in to the business object;

[0238] Based on the recommendation probability of the at least one target plug-in, a first recommended plug-in is selected from the at least one target plug-in.

[0239] In one implementation, the processing unit 802 is configured to select a first recommended plug-in from the at least one target plug-in based on the recommendation probability of the at least one target plug-in, and is specifically configured to perform the following steps:

[0240] Optimizing the recommendation probability of the at least one target plug-in based on the similarity between the plug-in requirement information and the plug-in document of the at least one target plug-in to obtain an optimized probability of the at least one target plug-in;

[0241] A first recommended plug-in is selected from the at least one target plug-in according to the optimization probability of the at least one target plug-in.

[0242] In one implementation, the processing unit 802 is configured to filter out at least one target plug-in document from the plug-in documents of the plurality of candidate plug-ins based on similarities between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins, and specifically to perform the following steps:

[0243] Get the software type of the target software for which the plug-in is to be installed;

[0244] Performing type matching processing on the software type of the target software and the plug-in service types of the multiple candidate plug-ins, and screening the plug-in documents of the multiple reference plug-ins from the plug-in documents of the multiple candidate plug-ins;

[0245] Based on the similarity between the plug-in requirement information and the plug-in documents of the multiple reference plug-ins, at least one plug-in document of the target plug-in is filtered out from the plug-in documents of the multiple reference plug-ins.

[0246] In one implementation, processing unit 802 is configured to perform semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching policy, and to select a first recommended plug-in from the at least one target plug-in by performing the following steps:

[0247] Adjust the plug-in requirement information according to the software description information of the target software to obtain the adjusted plug-in requirement information;

[0248] According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the adjusted plug-in requirement information and the plug-in document of at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

[0249] In one implementation, the processing unit 802 is configured to filter out at least one target plug-in document from the plug-in documents of the plurality of candidate plug-ins based on similarities between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins, and specifically to perform the following steps:

[0250] Get similarity adjustment factor;

[0251] Performing distribution adjustment processing on the similarities between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins according to the distribution adjustment direction corresponding to the similarity adjustment factor, thereby obtaining selection probabilities of the plug-in documents of the multiple candidate plug-ins;

[0252] Based on the selection probabilities of the plug-in documents of the multiple candidate plug-ins, at least one plug-in document of the target plug-in is screened out from the plug-in documents of the multiple candidate plug-ins.

[0253] In one implementation, processing unit 802 is configured to perform semantic matching between plug-in requirement information and plug-in documents of multiple candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and to select a first recommended plug-in from the multiple candidate plug-ins by specifically executing the following steps:

[0254] Perform semantic understanding on the plug-in requirement information to obtain the semantic features of the plug-in requirement information;

[0255] Perform semantic understanding on the plug-in document of each candidate plug-in to obtain the document semantic features corresponding to the plug-in document of each candidate plug-in;

[0256] Perform feature matching between the demand semantic features and the document semantic features corresponding to the plug-in document of each candidate plug-in, and predict the recommendation probability of each candidate plug-in to the business object;

[0257] A first recommended plug-in is selected from the multiple candidate plug-ins according to the recommendation probabilities of the multiple candidate plug-ins.

[0258] In one implementation, the processing unit 802 is configured to, when determining, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information, specifically perform the following steps:

[0259] Obtaining a recommended content type corresponding to the plug-in requirement information, where the recommended content type is preset, or the recommended content type is obtained by performing content recognition on the plug-in requirement information;

[0260] Performing content screening processing on the plug-in document of the first recommended plug-in according to the recommended content type to obtain screened document content;

[0261] A semantic matching process is performed between the plug-in requirement information and the filtered document content, and a first recommended content semantically related to the plug-in requirement information is determined from the filtered document content.

[0262] In one implementation, the processing unit 802 is further configured to perform the following steps:

[0263] updating the semantic matching process indicated by the plug-in matching strategy according to the recommendation feedback information to obtain an updated plug-in matching strategy;

[0264] Perform semantic matching between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins according to the semantic matching process indicated by the updated plug-in matching strategy, and select a second recommended plug-in from the multiple candidate plug-ins;

[0265] Determining, in the plug-in document of the second recommended plug-in, second recommended content semantically related to the plug-in requirement information;

[0266] Perform plug-in recommendation processing on the business object according to the second recommended plug-in and the second recommended content.

[0267] In one implementation, the training process of the information processing model includes:

[0268] Obtaining sample data for training an information processing model, the sample data including sample requirement information and plug-in documents of a plurality of sample plug-ins;

[0269] Calling the retrieval model to perform similarity matching between the sample requirement information and the plug-in documents of the multiple sample plug-ins, and obtaining the selection probability of the retrieval model for the plug-in documents of the multiple sample plug-ins;

[0270] Calling the language model to perform semantic matching between the sample requirement information and the plug-in documents of multiple sample plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and predicting the recommendation probability of the multiple sample plug-ins;

[0271] According to the difference between the selection probabilities of the plug-in documents of the multiple sample plug-ins and the recommendation probabilities of the multiple sample plug-ins, loss information of the information processing model is determined, and the retrieval model in the information processing model is trained according to the loss information.

[0272] According to one embodiment of the present application, Figure 8The various units in the information processing device shown can be individually or all combined into one or several other units to constitute, or one (some) of the units can also be split into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the functions of a unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the device may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

[0273] According to another embodiment of the present application, the program can be executed by running on a general computing device such as a computer including a central processing unit (CPU), a random access memory (RAM), a read-only memory (ROM) and other processing elements and storage elements. Figure 3 、 Figure 4 or Figure 6 A computer program for each step involved in part or all of the method shown is constructed as follows Figure 8 The information processing device shown in and the information processing method of the embodiment of the present application are implemented. The computer program can be recorded on, for example, a computer-readable storage medium, and loaded into the above-mentioned computing device through the computer-readable storage medium and run therein.

[0274] In an embodiment of the present application, plug-in requirement information of a business object can be obtained, and plug-in documents of multiple candidate plug-ins can be obtained, and a plug-in matching strategy can be obtained. The plug-in matching strategy can be used to indicate a semantic matching process between the plug-in requirement information and the plug-in document. According to the semantic matching process indicated by the plug-in matching strategy, semantic matching can be performed between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and a first recommended plug-in can be selected from the multiple candidate plug-ins. It can be seen that through semantic matching, the plug-in document of the selected first recommended plug-in can be semantically matched with the plug-in requirement information, that is, the plug-in document of the first recommended plug-in can better meet the needs of the business object, thereby improving the accuracy of plug-in recommendation by performing plug-in recommendation processing based on the first recommended plug-in that meets the needs of the business object. In addition, first recommended content semantically related to the plug-in requirement information can be determined in the plug-in document of the first recommended plug-in, so that the first recommended content is fully consistent with the plug-in requirement information, thereby improving the accuracy of plug-in recommendation by performing plug-in recommendation processing based on the first recommended content that meets the needs of the business object.

[0275] Based on the above method and device embodiments, the present application provides a computer device. Figure 9 , Figure 9It is a structural diagram of a computer device provided in an embodiment of the present application. Figure 9 The computer device shown includes at least a processor 901, an input interface 902, an output interface 903, and a computer-readable storage medium 904. The processor 901, the input interface 902, the output interface 903, and the computer-readable storage medium 904 may be connected via a bus or other means.

[0276] The computer-readable storage medium 904 can be stored in a memory of a computer device. The computer-readable storage medium 904 is used to store a computer program, which includes computer instructions. The processor 1401 is used to execute the computer program stored in the computer-readable storage medium 904. The processor 901 (or CPU (Central Processing Unit)) is the computing core and control core of the computer device. It is suitable for implementing computer programs, specifically loading and executing computer programs to implement corresponding method processes or corresponding functions.

[0277] The embodiment of the present application also provides a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the computer device. In addition, a computer program suitable for being loaded and executed by the processor is also stored in the storage space. It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory (Non-Volatile Memory), such as at least one disk memory; optionally, it can also be at least one computer-readable storage medium located away from the aforementioned processor.

[0278] The computer device may be a server. In a specific implementation, the processor 901 may load and execute a computer program stored in the computer-readable storage medium 904 to implement the above-mentioned Figure 3 、 Figure 4 or Figure 6 In a specific implementation, the computer program in the computer-readable storage medium 904 is loaded by the processor 901 and executes the following steps:

[0279] Obtain plug-in requirement information for business objects and plug-in documents for multiple candidate plug-ins;

[0280] Get the plug-in matching strategy, which is used to indicate the semantic matching process between the plug-in requirement information and the plug-in document;

[0281] Perform semantic matching between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and select a first recommended plug-in from the plurality of candidate plug-ins;

[0282] Determining, in a plug-in document of a first recommended plug-in, first recommended content semantically related to the plug-in requirement information;

[0283] Perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

[0284] In one implementation, the computer program in the computer-readable storage medium 904 is loaded by the processor 901 and executes a semantic matching process according to the plug-in matching strategy, performs semantic matching between plug-in requirement information and plug-in documents of multiple candidate plug-ins, and selects a first recommended plug-in from the multiple candidate plug-ins, specifically for performing the following steps:

[0285] Based on the similarity between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins, screening out at least one plug-in document of the target plug-in from the plug-in documents of the plurality of candidate plug-ins;

[0286] According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the plug-in requirement information and the plug-in document of at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

[0287] In one implementation, the computer program in the computer-readable storage medium 904 is loaded by the processor 901 and executes a semantic matching process according to the plug-in matching strategy, performs semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in, and is specifically configured to perform the following steps when selecting a first recommended plug-in from the at least one target plug-in:

[0288] Performing semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and predicting a probability of recommending at least one target plug-in to the business object;

[0289] Based on the recommendation probability of the at least one target plug-in, a first recommended plug-in is selected from the at least one target plug-in.

[0290] In one implementation, the computer program in the computer-readable storage medium 904 is loaded and executed by the processor 901 to select a first recommended plug-in from the at least one target plug-in based on the recommendation probability of the at least one target plug-in, specifically for performing the following steps:

[0291] Optimizing the recommendation probability of the at least one target plug-in based on the similarity between the plug-in requirement information and the plug-in document of the at least one target plug-in to obtain an optimized probability of the at least one target plug-in;

[0292] A first recommended plug-in is selected from the at least one target plug-in according to the optimization probability of the at least one target plug-in.

[0293] In one implementation, the computer program in the computer-readable storage medium 904 is loaded and executed by the processor 901 to filter out at least one target plug-in document from the plug-in documents of the multiple candidate plug-ins based on similarities between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins, and is specifically configured to perform the following steps:

[0294] Get the software type of the target software for which the plug-in is to be installed;

[0295] Performing type matching processing on the software type of the target software and the plug-in service types of the multiple candidate plug-ins, and screening the plug-in documents of the multiple reference plug-ins from the plug-in documents of the multiple candidate plug-ins;

[0296] Based on the similarity between the plug-in requirement information and the plug-in documents of the multiple reference plug-ins, at least one plug-in document of the target plug-in is filtered out from the plug-in documents of the multiple reference plug-ins.

[0297] In one implementation, the computer program in the computer-readable storage medium 904 is loaded by the processor 901 and executes a semantic matching process according to the plug-in matching strategy, performs semantic matching between the plug-in requirement information and the plug-in document of at least one target plug-in, and is specifically configured to perform the following steps when selecting a first recommended plug-in from the at least one target plug-in:

[0298] Adjust the plug-in requirement information according to the software description information of the target software to obtain the adjusted plug-in requirement information;

[0299] According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the adjusted plug-in requirement information and the plug-in document of at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

[0300] In one implementation, the computer program in the computer-readable storage medium 904 is loaded and executed by the processor 901 to filter out at least one target plug-in document from the plug-in documents of the multiple candidate plug-ins based on similarities between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins, and is specifically configured to perform the following steps:

[0301] Get similarity adjustment factor;

[0302] Performing distribution adjustment processing on the similarities between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins according to the distribution adjustment direction corresponding to the similarity adjustment factor, thereby obtaining selection probabilities of the plug-in documents of the multiple candidate plug-ins;

[0303] Based on the selection probabilities of the plug-in documents of the multiple candidate plug-ins, at least one plug-in document of the target plug-in is screened out from the plug-in documents of the multiple candidate plug-ins.

[0304] In one implementation, the computer program in the computer-readable storage medium 904 is loaded by the processor 901 and executes a semantic matching process according to the plug-in matching strategy, performs semantic matching between plug-in requirement information and plug-in documents of multiple candidate plug-ins, and selects a first recommended plug-in from the multiple candidate plug-ins, specifically for performing the following steps:

[0305] Perform semantic understanding on the plug-in requirement information to obtain the semantic features of the plug-in requirement information;

[0306] Perform semantic understanding on the plug-in document of each candidate plug-in to obtain the document semantic features corresponding to the plug-in document of each candidate plug-in;

[0307] Perform feature matching between the demand semantic features and the document semantic features corresponding to the plug-in document of each candidate plug-in, and predict the recommendation probability of each candidate plug-in to the business object;

[0308] A first recommended plug-in is selected from the multiple candidate plug-ins according to the recommendation probabilities of the multiple candidate plug-ins.

[0309] In one implementation, when the computer program in the computer-readable storage medium 904 is loaded and executed by the processor 901 and determines, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information, the computer program is specifically configured to perform the following steps:

[0310] Obtaining a recommended content type corresponding to the plug-in requirement information, where the recommended content type is preset, or the recommended content type is obtained by performing content recognition on the plug-in requirement information;

[0311] Performing content screening processing on the plug-in document of the first recommended plug-in according to the recommended content type to obtain screened document content;

[0312] A semantic matching process is performed between the plug-in requirement information and the filtered document content, and a first recommended content semantically related to the plug-in requirement information is determined from the filtered document content.

[0313] In one implementation, the computer program in the computer-readable storage medium 904 is loaded by the processor 901 and is further configured to perform the following steps:

[0314] updating the semantic matching process indicated by the plug-in matching strategy according to the recommendation feedback information to obtain an updated plug-in matching strategy;

[0315] Perform semantic matching between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins according to the semantic matching process indicated by the updated plug-in matching strategy, and select a second recommended plug-in from the multiple candidate plug-ins;

[0316] Determining, in the plug-in document of the second recommended plug-in, second recommended content semantically related to the plug-in requirement information;

[0317] Perform plug-in recommendation processing on the business object according to the second recommended plug-in and the second recommended content.

[0318] In one implementation, the training process of the information processing model includes:

[0319] Obtaining sample data for training an information processing model, the sample data including sample requirement information and plug-in documents of a plurality of sample plug-ins;

[0320] Calling the retrieval model to perform similarity matching between the sample requirement information and the plug-in documents of the multiple sample plug-ins, and obtaining the selection probability of the retrieval model for the plug-in documents of the multiple sample plug-ins;

[0321] Calling the language model to perform semantic matching between the sample requirement information and the plug-in documents of multiple sample plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and predicting the recommendation probability of the multiple sample plug-ins;

[0322] According to the difference between the selection probabilities of the plug-in documents of the multiple sample plug-ins and the recommendation probabilities of the multiple sample plug-ins, loss information of the information processing model is determined, and the retrieval model in the information processing model is trained according to the loss information.

[0323] In an embodiment of the present application, plug-in requirement information of a business object can be obtained, and plug-in documents of multiple candidate plug-ins can be obtained, and a plug-in matching strategy can be obtained. The plug-in matching strategy can be used to indicate a semantic matching process between the plug-in requirement information and the plug-in document. According to the semantic matching process indicated by the plug-in matching strategy, semantic matching can be performed between the plug-in requirement information and the plug-in documents of multiple candidate plug-ins, and a first recommended plug-in can be selected from the multiple candidate plug-ins. It can be seen that through semantic matching, the plug-in document of the selected first recommended plug-in can be semantically matched with the plug-in requirement information, that is, the plug-in document of the first recommended plug-in can better meet the needs of the business object, thereby improving the accuracy of plug-in recommendation by performing plug-in recommendation processing based on the first recommended plug-in that meets the needs of the business object. In addition, first recommended content semantically related to the plug-in requirement information can be determined in the plug-in document of the first recommended plug-in, so that the first recommended content is fully consistent with the plug-in requirement information, thereby improving the accuracy of plug-in recommendation by performing plug-in recommendation processing based on the first recommended content that meets the needs of the business object.

[0324] The present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the above-mentioned information processing method.

[0325] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0326] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0327] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrations. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0328] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An information processing method, characterized in that: include: Obtain plug-in requirement information for business objects and plug-in documents for multiple candidate plug-ins; Obtaining a plug-in matching strategy, where the plug-in matching strategy is used to indicate a semantic matching process between the plug-in requirement information and the plug-in document; performing semantic matching between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and selecting a first recommended plug-in from the plurality of candidate plug-ins; Determining, in a plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information; Perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

2. The method according to claim 1, wherein The step of performing semantic matching between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and selecting a first recommended plug-in from the plurality of candidate plug-ins, comprises: Based on the similarity between the plug-in requirement information and the plug-in documents of the multiple candidate plug-ins, screening out at least one target plug-in document from the plug-in documents of the multiple candidate plug-ins; According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the plug-in requirement information and the plug-in document of the at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

3. The method according to claim 2, wherein The step of performing semantic matching between the plug-in requirement information and the plug-in document of the at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and selecting a first recommended plug-in from the at least one target plug-in, includes: performing semantic matching between the plug-in requirement information and the plug-in document of the at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and predicting a probability of recommending the at least one target plug-in to the business object; Based on the recommendation probability of the at least one target plug-in, a first recommended plug-in is selected from the at least one target plug-in.

4. The method according to claim 3, wherein The selecting a first recommended plug-in from the at least one target plug-in based on the recommendation probability of the at least one target plug-in includes: optimizing the recommendation probability of the at least one target plug-in based on the similarity between the plug-in requirement information and the plug-in document of the at least one target plug-in to obtain an optimized probability of the at least one target plug-in; A first recommended plug-in is selected from the at least one target plug-in according to the optimization probability of the at least one target plug-in.

5. The method according to claim 2, wherein The step of selecting at least one target plug-in document from the plug-in documents of the plurality of candidate plug-ins based on the similarity between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins includes: Get the software type of the target software for which the plug-in is to be installed; Performing type matching processing on the software type of the target software and the plug-in service types of the multiple candidate plug-ins, and screening multiple reference plug-in documents from the plug-in documents of the multiple candidate plug-ins; Based on the similarity between the plug-in requirement information and the plug-in documents of the multiple reference plug-ins, at least one plug-in document of a target plug-in is screened out from the plug-in documents of the multiple reference plug-ins.

6. The method according to claim 5, wherein The step of performing semantic matching between the plug-in requirement information and the plug-in document of the at least one target plug-in according to the semantic matching process indicated by the plug-in matching strategy, and selecting a first recommended plug-in from the at least one target plug-in, includes: Performing demand adjustment processing on the plug-in requirement information according to the software description information of the target software to obtain adjusted plug-in requirement information; According to the semantic matching process indicated by the plug-in matching strategy, semantic matching is performed between the adjusted plug-in requirement information and the plug-in document of the at least one target plug-in, and a first recommended plug-in is selected from the at least one target plug-in.

7. The method according to claim 2, wherein The step of selecting at least one target plug-in document from the plug-in documents of the plurality of candidate plug-ins based on the similarity between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins includes: Get similarity adjustment factor; performing distribution adjustment processing on the similarities between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the distribution adjustment direction corresponding to the similarity adjustment factor, to obtain selection probabilities of the plug-in documents of the plurality of candidate plug-ins; Based on the selection probabilities of the plug-in documents of the multiple candidate plug-ins, at least one plug-in document of a target plug-in is screened out from the plug-in documents of the multiple candidate plug-ins.

8. The method according to claim 1, wherein The step of performing semantic matching between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and selecting a first recommended plug-in from the plurality of candidate plug-ins, comprises: Performing semantic understanding on the plug-in requirement information to obtain requirement semantic features corresponding to the plug-in requirement information; Performing semantic understanding on the plug-in document of each candidate plug-in to obtain document semantic features corresponding to the plug-in document of each candidate plug-in; Performing feature matching on the requirement semantic feature and the document semantic feature corresponding to the plug-in document of each candidate plug-in, and predicting a recommendation probability of recommending each candidate plug-in to the business object; A first recommended plug-in is selected from the multiple candidate plug-ins according to the recommendation probabilities of the multiple candidate plug-ins.

9. The method according to any one of claims 1 to 8, wherein The determining, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information includes: Obtaining a recommended content type corresponding to the plug-in requirement information, where the recommended content type is preset, or the recommended content type is obtained by performing content recognition on the plug-in requirement information; Performing content screening processing on the plug-in document of the first recommended plug-in according to the recommended content type to obtain screened document content; A semantic matching process is performed between the plug-in requirement information and the filtered document content, and a first recommended content semantically related to the plug-in requirement information is determined from the filtered document content.

10. The method according to any one of claims 1 to 8, wherein The method further comprises: updating the semantic matching process indicated by the plug-in matching strategy according to the recommendation feedback information to obtain an updated plug-in matching strategy; performing semantic matching between the plug-in requirement information and the plug-in documents of the plurality of candidate plug-ins according to the semantic matching process indicated by the updated plug-in matching strategy, and selecting a second recommended plug-in from the plurality of candidate plug-ins; determining, in a plug-in document of the second recommended plug-in, second recommended content semantically related to the plug-in requirement information; Perform plug-in recommendation processing on the business object according to the second recommended plug-in and the second recommended content.

11. The method according to claim 1, wherein The method is performed by an information processing model, which includes a retrieval model and a language model. The training process of the information processing model includes: Acquiring sample data for training the information processing model, the sample data including sample requirement information and plug-in documents of a plurality of sample plug-ins; calling the retrieval model to perform similarity matching between the sample requirement information and the plug-in documents of the plurality of sample plug-ins, and obtaining a selection probability of the retrieval model for the plug-in documents of the plurality of sample plug-ins; Calling the language model to perform semantic matching between the sample requirement information and the plug-in documents of the multiple sample plug-ins according to the semantic matching process indicated by the plug-in matching strategy, and predicting recommendation probabilities for the multiple sample plug-ins; According to the difference between the selection probabilities of the plug-in documents of the multiple sample plug-ins and the recommendation probabilities of the multiple sample plug-ins, loss information of the information processing model is determined, and the retrieval model in the information processing model is trained according to the loss information.

12. An information processing device, characterized in that: include: An acquisition unit, used to acquire plug-in requirement information of a business object and to acquire plug-in documents of multiple candidate plug-ins; The acquisition unit is further configured to acquire a plug-in matching strategy, where the plug-in matching strategy is configured to indicate a semantic matching process between plug-in requirement information and plug-in documents; a processing unit, configured to perform semantic matching between the plug-in requirement information and plug-in documents of a plurality of candidate plug-ins according to a semantic matching process indicated by the plug-in matching strategy, and select a first recommended plug-in from the plurality of candidate plug-ins; The processing unit is further configured to determine, in the plug-in document of the first recommended plug-in, first recommended content semantically related to the plug-in requirement information; The processing unit is further configured to perform plug-in recommendation processing on the business object according to the first recommended plug-in and the first recommended content.

13. A computer device, characterized in that: The computer device comprises: a processor suitable for implementing a computer program; A computer-readable storage medium storing a computer program, wherein the computer program is suitable for being loaded by the processor and executing the information processing method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the information processing method according to any one of claims 1 to 11.

15. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the information processing method according to any one of claims 1 to 11 is implemented.

Images