Resource matching method, person and post matching method and corresponding devices

Abstract

The invention provides a resource matching method, a person and post matching method and corresponding devices, and the resource matching method comprises the steps: obtaining a description text of a first object, a description text of a second object, a historical interaction sequence of the first object and a historical interaction sequence of the second object, the historical interaction sequence of the first object comprises object information of at least one second type resource generating preset type interaction with the first object, and the historical interaction sequence of the second object comprises object information of at least one first type resource generating preset type interaction with the second object; and encoding the description text of the first object, the description text of the second object, the historical interaction sequence of the first object and the historical interaction sequence of the second object, and determining a matching result between the first object and the second object by using feature representation obtained by encoding. According to the invention, the resource matching accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer applications, and in particular to a resource matching method, a person-job matching method, and corresponding devices. Background Art

[0002] With the rapid development of computer technology, more and more services are being implemented through online service systems. In some scenarios, these systems involve resource matching. For example, online recruitment systems can efficiently match candidates with positions through online communication between job seekers and employers. However, current online service systems such as online recruitment systems still face some challenges, particularly poor accuracy in resource matching. Summary of the Invention

[0003] In view of this, the present application provides a resource matching method, a person-job matching method and corresponding devices to improve the accuracy of resource matching.

[0004] This application provides the following solutions:

[0005] In a first aspect, a resource matching method is provided, including: obtaining a description text of a first object, a description text of a second object, a historical interaction sequence of the first object, and a historical interaction sequence of the second object, wherein the first object and the second object belong to a first type of resource and a second type of resource, respectively, the historical interaction sequence of the first object includes object information of at least one second type of resource that generates a preset type of interaction with the first object, and the historical interaction sequence of the second object includes object information of at least one first type of resource that generates a preset type of interaction with the second object; encoding the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object, and determining a matching result between the first object and the second object using a feature representation obtained by encoding.

[0006] In some embodiments, the description text of the second object is obtained by enhancing the original description text of the second object using the large language model, wherein enhancing the original description text of the second object using the large language model includes:

[0007] If the quality of the original description text of the second object is lower than the preset quality requirement, construct a prompt instruction using the original description text of the second object, the enhanced task information, and the thought chain prompt template, input the prompt instruction into the large language model, and obtain the enhanced text output by the large language model as the description text of the second object;

[0008] The thought chain prompt template includes an inference path for instructing the large language model to perform the enhancement.

[0009] In some embodiments, the reasoning path includes: analyzing the original description text of the second object to obtain attribute information corresponding to the second object; determining whether the second object has object information of the first type of resource that has been historically matched successfully; if so, extracting common information from the object information of the first type of resource that has been historically matched successfully, and using the common information and attribute information to enhance the original description text of the second object to obtain the description text of the second object; if not, using the attribute information to enhance the original description text of the second object to obtain the description text of the second object.

[0010] In some embodiments, encoding the descriptive text of the first object, the descriptive text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object includes: respectively obtaining the embedded representation corresponding to the descriptive text of the first object, the embedded representation corresponding to the descriptive text of the second object, the embedded representation corresponding to the historical interaction sequence of the first object, and the embedded representation corresponding to the historical interaction sequence of the second object; using a first encoder to respectively encode the embedded representation corresponding to the descriptive text of the first object and the embedded representation corresponding to the historical interaction sequence of the first object to obtain the feature representation corresponding to the descriptive text of the first object and the feature representation corresponding to the historical interaction sequence of the first object; and using a second encoder to respectively encode the embedded representation corresponding to the descriptive text of the second object and the embedded representation corresponding to the historical interaction sequence of the second object to obtain the feature representation corresponding to the descriptive text of the second object and the feature representation corresponding to the historical interaction sequence of the second object.

[0011] In some embodiments, both the first encoder and the second encoder include an encoding module; encoding the embedded representation corresponding to the description text includes: using the encoding module to encode the embedded representation corresponding to the description text to obtain a feature representation corresponding to the description text.

[0012] In some embodiments, both the first encoder and the second encoder include an encoding module and a fusion module; encoding the embedded representation corresponding to the historical interaction sequence includes: using the fusion module to perform multi-head attention processing on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the first object to obtain a first fused representation; performing multi-head attention processing on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the second object to obtain a second fused representation; splicing the first fused representation and the second fused representation to obtain a third fused representation; using the encoding module to encode the third fused representation to obtain a feature representation corresponding to the historical interaction sequence.

[0013] In some embodiments, the embedded representation is obtained by an embedding network, and the embedding network is pre-trained in the following manner:

[0014] Obtaining a sample pair including a positive sample and a negative sample, wherein the positive sample includes sample data of a second object sample and sample data of a first object sample that matches the second object sample, and the negative sample includes sample data of the second object sample and sample data of the first object sample that does not match the second object sample, and the sample data includes a description text and a historical interaction sequence;

[0015] The general vector model is used as the initial model of the embedding network, and the initial model is trained using the sample pairs in a contrastive learning manner to obtain the embedding network.

[0016] In some embodiments, determining the matching result between the first object and the second object using the feature representation obtained by encoding includes: using multiple expert networks in the expert mixture network to predict the matching value between the first object and the second object based on the feature representation obtained by encoding; and performing weighted summation on the matching values predicted by the multiple expert networks to obtain the matching result between the first object and the second object.

[0017] In some embodiments, determining the matching result between the first object and the second object using the encoded feature representation also includes: using a gating network based on a pre-trained type embedding representation to determine the weights corresponding to multiple expert networks; and based on the weights corresponding to the multiple expert networks, performing a step of weighted summing of the matching values predicted by the multiple expert networks.

[0018] On the second aspect, a person-job matching method is provided, including: obtaining a description text of a target candidate, a description text of a target job, a historical interaction sequence of the target candidate, and a historical interaction sequence of the target job, the historical interaction sequence of the target candidate including at least one job information that generates a preset type of interaction with the target candidate, and the historical interaction sequence of the target job including at least one candidate information that generates a preset type of interaction with the target job; encoding the description text of the target candidate, the description text of the target job, the historical interaction sequence of the target candidate, and the historical interaction sequence of the target job, and using the feature representation obtained by encoding to determine the matching result between the target candidate and the target job.

[0019] In a third aspect, a resource matching device is provided, including: an acquisition module for acquiring a description text of a first object, a description text of a second object, a historical interaction sequence of the first object, and a historical interaction sequence of the second object, wherein the first object and the second object belong to a first type of resource and a second type of resource, respectively, the historical interaction sequence of the first object includes object information of at least one second type of resource that generates a preset type of interaction with the first object, and the historical interaction sequence of the second object includes object information of at least one first type of resource that generates a preset type of interaction with the second object; a matching module for encoding the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object, and determining a matching result between the first object and the second object using the feature representation obtained by encoding.

[0020] In a fourth aspect, a person-job matching device is provided, including: an acquisition module for acquiring a description text of a target candidate, a description text of a target job, a historical interaction sequence of the target candidate, and a historical interaction sequence of the target job, wherein the historical interaction sequence of the target candidate includes at least one job information that generates a preset type of interaction with the target candidate, and the historical interaction sequence of the target job includes at least one candidate information that generates a preset type of interaction with the target job; a matching module for encoding the description text of the target candidate, the description text of the target job, the historical interaction sequence of the target candidate, and the historical interaction sequence of the target job, and determining the matching result between the target candidate and the target job using the feature representation obtained by encoding.

[0021] In a fifth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed, the steps of the method of the first aspect or the second aspect are implemented.

[0022] In a sixth aspect, an electronic device is provided, comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, execute the steps of the method of the first aspect or the second aspect.

[0023] In a seventh aspect, a computer program product is provided, comprising a computer program, which implements the steps of the method of the first aspect or the second aspect when the computer program is executed.

[0024] The resource matching method, person-job matching method, device, and electronic equipment provided in this application can achieve the following technical effects:

[0025] 1) When matching first-type resources and second-type resources, this application, in addition to considering the descriptive text of the first object and the second object in the two types of resources, further considers the historical interaction sequence of the first object and the historical interaction sequence of the second object, that is, introduces two-way historical interaction information into resource matching, and refines the historical interaction sequence through preset type interactions to improve the refinement of the feature representation corresponding to the historical interaction sequence, thereby improving the matching accuracy between the first object and the second object.

[0026] 2) This application introduces a large language model to perform data enhancement processing on the short original description text of the second object, which can improve the quality of the description text of the second object, thereby improving the quality of the feature representation obtained after encoding the description text of the second object, and further improving the accuracy of determining the matching results between the first object and the second object.

[0027] 3) This application designs a thinking chain prompt template to guide the large language model to enhance the original description text of the second object according to the constrained reasoning path, thereby improving the text enhancement quality of the large language model and further improving the accuracy of resource matching.

[0028] 4) In this application, when encoding the embedded representation corresponding to the historical interaction sequence, the description text of the first object and the description text of the second object are integrated into the feature representation corresponding to the historical interaction sequence based on multi-head attention processing, so that the model can capture more comprehensive interaction information, thereby improving the accuracy of resource matching.

[0029] 5) This application uses object samples of the first type of resources and the second type of resources to construct positive and negative sample pairs, and adopts contrastive learning to allow the embedding network to learn the specific semantic relationship between the two types of resources in the field, thereby improving the representation ability of the embedding network in this field and further improving the accuracy of resource matching.

[0030] 6) This application introduces an expert hybrid network, so that each expert network is responsible for learning the matching relationship between the first object and the second object from a different perspective. Each expert network uses the feature representation obtained after encoding to predict the matching degree between the first object and the second object, which can reduce the impact of inconsistent judgment criteria in different scenarios, thereby improving the accuracy of resource matching.

[0031] 7) This application introduces type embedding representation into the expert mixture network, and uses a gating network to determine the weight corresponding to each expert network in the expert mixture network based on the type embedding representation obtained in advance through training. That is, the type embedding representation is used as a guide for adaptive expert network routing, so that the expert mixture network has scene perception capabilities, and then a weighted summation method is used to determine the matching degree between the first object and the second object, which can further improve the accuracy and robustness of resource matching.

[0032] 8) When this application is applied to person-job matching, in addition to considering the descriptive texts of the target candidate and the target job, it further considers the historical interaction sequences of the target candidate and the historical interaction sequences of the target job, introduces two-way person-job interaction information into person-job matching, and refines the historical interaction sequences through preset type interactions, thereby improving the refinement of the feature representation corresponding to the historical interaction sequences, thereby improving the accuracy of person-job matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] 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. 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 creative work.

[0034] Figure 1 A schematic diagram of the system architecture provided in an embodiment of the present application;

[0035] Figure 2 A schematic diagram of a resource matching method according to an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of a resource matching model provided in an embodiment of the present application;

[0037] Figure 4 A schematic diagram of a first encoder provided in an embodiment of the present application;

[0038] Figure 5 A flowchart of the person-job matching method provided in an embodiment of the present application;

[0039] Figure 6 A schematic diagram of the structure of a resource matching device provided in an embodiment of the present application;

[0040] Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] 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.

[0042] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0043] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0044] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0045] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0046] With the rapid development of internet technology, the inventors have discovered through research that current matching systems face challenges in determining the degree of match between two objects of different resource types, resulting in low accuracy. For example, in assessing the degree of match between candidates and positions, many online recruitment systems contain a large amount of low-quality job descriptions. These descriptions have low information density and are brief, which directly affects the accuracy of online recruitment systems in matching candidates. Furthermore, while some methods attempt to capture dynamic candidate preference information through historical application interaction data and candidate behavior sequences to improve the degree of match between candidates and positions, the problem of low accuracy in determining the degree of match between candidates and positions still exists.

[0047] In view of this, this application provides a new idea. Figure 1 This is a flow chart of the resource matching method provided in the embodiment of the present application. Figure 1 As shown in , the method may include the following steps:

[0048] Step 101: Obtain a description text of a first object, a description text of a second object, a historical interaction sequence of the first object, and a historical interaction sequence of the second object, wherein the first object and the second object belong to a first type of resource and a second type of resource, respectively, the historical interaction sequence of the first object includes object information of at least one second type of resource that generates a preset type of interaction with the first object, and the historical interaction sequence of the second object includes object information of at least one first type of resource that generates a preset type of interaction with the second object.

[0049] Step 103: Encode the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object, and use the encoded feature representation to determine the matching result between the first object and the second object.

[0050] It can be seen from the above process that when matching first-type resources and second-type resources, the present application not only considers the descriptive text of the first object and the second object in the two types of resources, but also further considers the historical interaction sequence of the first object and the historical interaction sequence of the second object, that is, introduces two-way historical interaction information into resource matching, and refines the historical interaction sequence through preset type interactions to improve the refinement of the feature representation corresponding to the historical interaction sequence, thereby improving the matching accuracy between the first object and the second object.

[0051] In order to facilitate understanding of the embodiments of the present application, the system architecture of the present application is introduced below. It should be noted that the system architecture is only an illustrative example and is not intended to be a specific description of the system architecture.

[0052] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the system architecture provided in the embodiment of the present application. Figure 2 As shown, the system architecture includes a user terminal and a server.

[0053] User terminals may include, but are not limited to, smart mobile terminals, smart home devices, wearable devices, and personal computers (PCs). Smart mobile devices may include mobile phones, tablets, laptops, PDAs (Personal Digital Assistants), and internet-connected cars. Smart home devices may include smart TVs and smart refrigerators. Wearable devices may include smart watches, smart glasses, virtual reality devices, augmented reality devices, and mixed reality devices (i.e., devices that support both virtual reality and augmented reality).

[0054] A server can be a standalone server, a server cluster, or even a cloud server. A cloud server, also known as a cloud computing server or cloud host, is a hosting product within the cloud computing service ecosystem. It addresses the management difficulties and limited scalability of traditional physical hosting and virtual private server (VPS) services.

[0055] The resource matching method provided in the embodiment of the present application can be executed on the server side or by a computer terminal with strong computing capabilities.

[0056] As one of the feasible methods, a user may input identification information of the first object and the second object through a user terminal, and the user terminal may send the identification information of the first object and the second object to a server. The server may match the first object and the second object using the method provided in an embodiment of the present application, and return the matching result to the user terminal, or return the resources obtained based on the matching result to the user terminal.

[0057] As another feasible method, the user may obtain the description text of the first object and the description text of the second object through the user terminal and send them to the server. The server matches the first object and the second object using the method provided in the embodiment of the present application and returns the matching result to the user terminal, or returns the resources obtained based on the matching result to the user terminal.

[0058] It should be understood that Figure 1 The number of user terminals and servers in the embodiment is merely illustrative. Any number of user terminals and servers may be provided as required.

[0059] The following describes each step in the above process in detail with reference to an embodiment. It should be noted that the terms "first" and "second" in this application do not restrict size, order, or quantity, but are merely used to distinguish objects by name. For example, "first object" and "second object" are used to distinguish two objects by name. Another example is "first type resource" and "second type resource" are used to distinguish two types of resources by name, and so on.

[0060] First, in conjunction with the embodiment, the above step 101, namely, “obtaining the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object”, is described in detail.

[0061] Among them, the first object and the second object belong to the first type of resources and the second type of resources respectively, the historical interaction sequence of the first object includes the object information of at least one second type of resource that generates a preset type of interaction with the first object, and the historical interaction sequence of the second object includes the object information of at least one first type of resource that generates a preset type of interaction with the second object.

[0062] For example, if the first object is a candidate and the second object is a position, the description text of the first object is the candidate's resume information, the description text of the second object is the position description information, the historical interaction sequence of the first object includes information about at least one position that had a preset type of interaction with the candidate, and the historical interaction sequence of the second object includes information about at least one candidate that had a preset type of interaction with the position. A matching result between the first and second objects is determined using the method provided in the embodiments of the present application, and based on the matching result, a position is recommended to the candidate, or a candidate is recommended for the position.

[0063] For another example, if the first object is a consumer and the second object is a product, the description text of the first object is the consumer's portrait information, the description text of the second object is the product description information, the first object's historical interaction sequence includes information about at least one product that had a preset type of interaction with the consumer, and the second object's historical interaction sequence includes information about at least one consumer that had a preset type of interaction with the product. A matching result between the first and second objects is determined using the methods provided in the embodiments of the present application, and products are recommended to the consumer based on the matching result.

[0064] For another example, if the first object is a user participating in an activity and the second object is an activity, then the description text of the first object is the profile information of the user participating in the activity, and the description text of the second object is the description information of the activity. The historical interaction sequence of the first object includes information about at least one activity that generated a preset type of interaction with the user participating in the activity, and the historical interaction sequence of the second object includes information about at least one user participating in the activity that generated a preset type of interaction with the activity. The matching result between the first object and the second object is determined using the method provided in the embodiments of the present application, and an activity is recommended to the user based on the matching result.

[0065] For another example, if the first object is a tenant and the second object is a house, then the description text of the first object is the tenant's portrait information, the description text of the second object is the house description information, the historical interaction sequence of the first object includes information about at least one house that had a preset type of interaction with the tenant, and the historical interaction sequence of the second object includes information about at least one tenant that had a preset type of interaction with the house. The matching result between the first and second objects is determined using the method provided in the embodiments of the present application, and a house is recommended to the tenant based on the matching result, or user information matching the house is provided to the house agent.

[0066] For another example, if the first object is a supplier and the second object is a project, the description text of the first object is the supplier's profile information, the description text of the second object is the project description information, the historical interaction sequence of the first object includes information about at least one project that had a preset type of interaction with the supplier, and the historical interaction sequence of the second object includes information about at least one supplier that had a preset type of interaction with the project. A matching result between the first and second objects is determined using the method provided in the embodiments of the present application, and suppliers are recommended to project managers based on the matching result.

[0067] In addition to the resource matching scenarios exemplified above, this application can also be used for matching scenarios between recipients and donated resources, matching scenarios between students and schools, matching scenarios between patients and medical resources, etc. This application does not limit the resource types of the first object and the second object.

[0068] The aforementioned preset interaction types can be used to categorize historical interaction sequences based on the degree of match between the first object and the second object. For example, based on the degree of match between the consumer and the product, historical interaction sequences can be categorized into types such as the consumer viewing product information, the consumer adding the product to the shopping cart, the consumer purchasing the product, and the consumer returning the product. For another example, based on the degree of match between the tenant and the property, historical interaction sequences can be categorized into types such as the tenant inspecting the property, the tenant signing the contract with the property, the tenant moving into the property, and the duration of the tenant's stay.

[0069] For another example, when matching people with jobs, the candidate's historical interaction sequence includes: job information of the candidate's resume being evaluated, job information of the candidate's resume being evaluated and passed, and job information of the candidate being interviewed and passed; the job's historical interaction sequence includes: candidate information of the resume being evaluated, candidate information of the resume being evaluated and passed, and candidate information of the candidate being interviewed and passed.

[0070] As one possible implementation manner, the description text of the second object may be the original description text.

[0071] However, some of the original descriptions of the second object are often of low quality and contain little useful information, which greatly complicates subsequent resource matching and results in inaccurate matching results. In view of this, the present application provides another more preferred implementation method, in which the description of the second object is obtained by enhancing the original description of the second object using a large language model.

[0072] The use of the large language model to enhance the original description text of the second object may include: if the quality of the original description text of the second object is lower than the preset quality requirement, using the original description text of the second object, enhanced task information and a chain of thought (COT) prompt template to construct a prompt instruction (Prompt), inputting the prompt instruction into the large language model, and obtaining the enhanced text output by the large language model as the description text of the second object.

[0073] The quality of the original description text can be measured by length. For example, an original description text whose length is less than or equal to a preset length threshold is considered to be below a preset quality requirement. In addition to length, other indicators such as information density can also be used to measure the quality of the original description text.

[0074] A large language model (LLM) refers to a deep learning model trained using large amounts of text data that can generate natural language text or understand the meaning of language text. It is characterized by its large scale and huge number of parameters (usually reaching over 10 billion levels), and is usually based on a deep learning architecture such as the Transformer architecture. The difference between an LLM and an ordinary pre-trained language model lies in the parameter scale. When the parameter scale exceeds a certain level, the model achieves significant performance improvements and exhibits capabilities that do not exist in small models, such as in-context learning capabilities, which can learn complex patterns in language and perform a wide range of tasks, including text summarization, translation, sentiment analysis, multi-round dialogue, and so on. Therefore, in order to distinguish it from traditional pre-trained language models, models with a parameter scale exceeding a certain level are called LLMs. In general, a language model with a parameter scale of over 10 billion based on a deep learning architecture can be considered a large language model.

[0075] Among them, the thought chain prompt template may include an inference path for instructing the large language model to perform enhancement. As one of the feasible ways, the inference path may include: analyzing the original description text of the second object to obtain the attribute information corresponding to the second object (for example, the skill requirements and industry fields corresponding to the position, the functions and usage fields of the goods, the location information, area, furniture configuration and rental conditions of the house, etc.); determining whether the second object has object information of the first type of resource that has been successfully matched historically; if so, extracting the common information in the object information of the first type of resource that has been successfully matched historically, and using the common information and attribute information to enhance the original description text of the second object (in this case, the domain knowledge of the resource is fully borrowed) to obtain the description text of the second object; if not, using the attribute information to enhance the original description text of the second object (mainly polishing and expanding) to obtain the description text of the second object.

[0076] In this embodiment, for the first object, in order to match the desired second object, the quality of its description text is relatively high, while the matched second object usually lacks the more critical description information of the second object due to information updates or in order to attract the first object. For example, in the matching scenario of candidates and positions, in order to fully demonstrate their competitiveness, candidates usually include detailed information such as educational background, work experience, skill certificates, project experience, etc. in their resumes, which has a high information density; while when publishers post positions, in order to attract more candidates, their job descriptions usually contain more information such as salary and work environment, but less information such as the specific work content of the position and the information required of candidates, resulting in lower quality of the job description information they post; or, when publishers post positions, they are under time pressure, resulting in very little content in the job description information, which leads to lower quality of the job description information they post.

[0077] In order to improve the quality of the original description text of the second object and to improve the accuracy of determining the matching degree between the first object and the second object, a large language model may be introduced to perform enhancement processing on the original description text of the second object. Figure 3 A schematic diagram of a resource matching model provided in an embodiment of the present application. Figure 3 , the resource matching model includes a large language model.

[0078] A thought chain prompt template can be designed for the large language model to guide it in generating high-quality descriptive text. The COT prompt template provides the large language model with its role and context, instructs it to perform data augmentation tasks, provides it with the specific solution process for the task, and constrains the output data of the large language model.

[0079] The large language model outputs a description of the second object according to the constraints of the COT prompt template. The constraints may include, but are not limited to, at least one of the following: retaining 70% of the keywords in the original description of the second object; reducing complex and ambiguous descriptions; and not making unwarranted expansions and / or rewording.

[0080] By guiding the large language model to perform the task according to the specific solution process of the task through the COT prompt template and constraining the output of the large language model through constraint conditions, the hallucination of the large language model can be reduced, and the noise in the output data of the large language model can be reduced, thereby improving the output quality of the large language model and further improving the accuracy of determining the matching result between the first object and the second object.

[0081] The following describes in detail "encoding the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object" in step 203 with reference to an embodiment.

[0082] In one embodiment, encoding the descriptive text of the first object, the descriptive text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object includes: respectively obtaining the embedded representation corresponding to the descriptive text of the first object, the embedded representation corresponding to the descriptive text of the second object, the embedded representation corresponding to the historical interaction sequence of the first object, and the embedded representation corresponding to the historical interaction sequence of the second object; using a first encoder to respectively encode the embedded representation corresponding to the descriptive text of the first object and the embedded representation corresponding to the historical interaction sequence of the first object to obtain a feature representation corresponding to the descriptive text of the first object and a feature representation corresponding to the historical interaction sequence of the first object; and using a second encoder to respectively encode the embedded representation corresponding to the descriptive text of the second object and the embedded representation corresponding to the historical interaction sequence of the second object to obtain a feature representation corresponding to the descriptive text of the second object and a feature representation corresponding to the historical interaction sequence of the second object.

[0083] Continue to refer Figure 3 First, the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object are converted into corresponding embedding representations through the embedding network to reduce the complexity of subsequent encoding. Figure 3 In

[15] , it is assumed that the historical interaction sequence of the first object includes object information of three types of interaction-type second-type resources, represented as sequence a, sequence b, and sequence c respectively; the historical interaction sequence of the second object includes object information of three types of interaction-type first-type resources, represented as sequence A, sequence B, and sequence C respectively. The embedding network outputs an embedding representation corresponding to the description text of the first object, three embedding representations corresponding to the historical interaction sequences of the first object (i.e., the embedding representation corresponding to sequence a, the embedding representation corresponding to sequence b, and the embedding representation corresponding to sequence c), an embedding representation corresponding to the description text of the second object, and three embedding representations corresponding to the historical interaction sequences of the second object (i.e., the embedding representation corresponding to sequence A, the embedding representation corresponding to sequence B, and the embedding representation corresponding to sequence C).

[0084] As one possible implementation, the above embedding network can use a general vector model. However, the general vector model does not have strong domain knowledge. Therefore, this application provides a more preferred embedding network, which is pre-trained in the following way:

[0085] A sample pair including positive samples and negative samples is obtained, where the positive sample includes sample data of the second object sample and sample data of the first object sample that matches the second object sample, and the negative sample includes sample data of the second object sample and sample data of the first object sample that does not match the second object sample, and the sample data includes descriptive text and historical interaction sequences; a universal vector model is used as an initial model of the embedding network, and the initial model is trained using the sample pairs by adopting a contrastive learning method to obtain the embedding network.

[0086] For the first type of resources and the second type of resources, the first object that matches the same object under the second type of resources and the first object that does not match can be collected to form positive and negative sample pairs respectively as training samples. The training sample can be a triplet training data, that is, training data composed of "second object-first object (+)-first object (-)", where the first object (+) is the first object that matches the second object and constitutes a positive sample with the second object; the first object (-) is the first object that does not match the second object and constitutes a negative sample with the second object. However, in the sample data corresponding to the first object and the second object that do not match under the second type of resources, there may be cases where the first object and the second object are not matched due to non-objective factors such as willingness. For example, although the candidate and the position are not successfully matched in the end (for example, they are not hired), it may be due to the candidate's personal willingness. These noise data may cause the trained embedding network to have low accuracy and poor robustness. Therefore, it is necessary to clean these noise data in the training samples.

[0087] When training the initial model using contrastive learning, the training objective is used to design a loss function. In each iteration, the value of the loss function is used to update the model parameters using methods such as gradient descent until the preset training termination conditions are met. The training termination conditions may include, for example, the value of the loss function being less than or equal to a preset loss function threshold, or the number of iterations reaching a preset threshold.

[0088] The training objectives are to minimize the distance between the embedding representations corresponding to the first object sample and the second object sample in the positive sample output by the embedding network, and to maximize the distance between the embedding representations corresponding to the first object sample and the second object sample in the negative sample output by the embedding network.

[0089] This application uses object samples of the first type of resources and the second type of resources to construct positive and negative sample pairs, and adopts contrastive learning to allow the embedding network to learn the specific semantic relationship between the fields where the two types of resources are located, thereby improving the representation ability of the embedding network in this field and further improving the accuracy of resource matching.

[0090] Continue to refer Figure 3, the resource matching model provided by the embodiment of the present application includes a first encoder, a second encoder and an expert mixture network. The embedded representation corresponding to the descriptive text of the first object output by the embedding network and the embedded representation corresponding to the historical interaction sequence of the first object can be input into the first encoder for encoding processing to obtain the feature representation corresponding to the descriptive text of the first object and the feature representation corresponding to the historical interaction sequence of the first object. The embedded representation corresponding to the descriptive text of the second object output by the embedding network and the embedded representation corresponding to the historical interaction sequence of the second object can be input into the second encoder for encoding processing to obtain the feature representation corresponding to the descriptive text of the second object and the feature representation corresponding to the historical interaction sequence of the second object.

[0091] In one embodiment, the first encoder and the second encoder both include an encoding module and a fusion module. The first encoder and the second encoder both involve encoding the description text. Taking the first encoder as an example, Figure 4 As shown in , the encoding module directly encodes the embedding representation corresponding to the description text of the first object to obtain the feature representation corresponding to the description text of the first object. The second encoder is similar to the first encoder, and the embedding representation corresponding to the second object is directly encoded by the encoding module to obtain the feature representation corresponding to the description text of the second object.

[0092] Among them, the encoding module can adopt a deep neural network (Deep Neural Network, DNN for short) and the like.

[0093] Both the first encoder and the second encoder are involved in encoding the embedded representation corresponding to the historical interaction sequence. Encoding the embedded representation corresponding to the historical interaction sequence may include: using a fusion module to perform multi-head attention processing on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the first object to obtain a first fused representation; performing multi-head attention processing on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the second object to obtain a second fused representation; concatenating the first fused representation and the second fused representation to obtain a third fused representation; and encoding the third fused representation using the encoding module to obtain a feature representation corresponding to the historical interaction sequence.

[0094] Taking the embedding representation corresponding to the first encoder processing sequence a as an example, Figure 4 Schematic diagram of the first encoder provided in the embodiment of the present application. Figure 4The fusion module performs multi-head attention processing on the embedding representation corresponding to the description text of the first object and the embedding representation corresponding to sequence a to obtain a first vector, and then performs a product operation on the first vector and the embedding representation corresponding to sequence a to obtain a first fused representation; and, performs multi-head attention processing on the embedding representation corresponding to the description text of the second object and the embedding representation corresponding to sequence a to obtain a second vector, and then performs a product operation on the second vector and the embedding representation corresponding to sequence a to obtain a second fused representation. After obtaining the first fused representation and the second fused representation, the fusion module splices the first fused representation and the second fused representation, and uses a network such as a DNN to transform the spliced vector to obtain a third fused representation. The encoding module encodes the third fused representation output by the fusion module to obtain the feature representation corresponding to sequence a.

[0095] The first encoder processes the embedding representations corresponding to sequence b and sequence c in the same way as it processes the embedding representations corresponding to sequence a, and will not be repeated here. Furthermore, the second encoder processes the embedding representations corresponding to sequence A, sequence B, and sequence C in the same way as the first encoder processes the embedding representations corresponding to sequence a, and will not be repeated here.

[0096] Because the information in historical interaction sequences is relatively complex, when encoding the embedded representation corresponding to the historical interaction sequence, a multi-head attention process is used to fuse the description text of the first object and the description text of the second object into the feature representation corresponding to the historical interaction sequence. This captures more comprehensive interaction information and thus extracts a more accurate feature representation corresponding to the historical interaction sequence. However, the information in the description text is relatively simple compared to the information in the historical interaction sequence. To improve encoding efficiency, the encoding module can be used to directly encode the embedded representation corresponding to the description text to obtain the corresponding feature representation of the description text.

[0097] The following describes in detail "determining the matching result between the first object and the second object using the feature representation obtained by encoding" in step 103 with reference to an embodiment.

[0098] As one of the feasible methods, multiple expert networks in the expert mixture network can be used to predict the matching value between the first object and the second object based on the feature representation obtained by encoding; the matching values predicted by the multiple expert networks are weighted and summed to obtain the matching result between the first object and the second object.

[0099] Continue to refer Figure 3 , Figure 3 The resource matching model shown also includes a mixture of experts network, which includes multiple expert networks (in Figure 3In the figure, expert network 1, expert network 2, ..., expert network n) are shown, each expert network can predict the matching value between the first object and the second object from a different perspective. Figure 3 In the method, the feature representation corresponding to the description text of a first object and the feature representation corresponding to the three sequences of the first object output by the first encoder, as well as the feature representation corresponding to the description text of a second object and the feature representation corresponding to the three sequences of the second object output by the second encoder are all input into the expert mixture network. Each expert network in the expert mixture network uses the eight input feature representations to perform inference and prediction to determine the matching value between the first object and the second object. Finally, the matching values predicted by each expert network are weighted and summed to obtain the final matching value between the first object and the second object.

[0100] A matching threshold can be set in advance. If the matching value output by the expert mixture network is greater than or equal to the matching threshold, the matching result between the first object and the second object is that the first object and the second object match. If the matching value output by the expert mixture network is less than the matching threshold, the matching result between the first object and the second object is that the first object and the second object do not match.

[0101] Optionally, the above-mentioned different angles are obtained by pre-training the expert mixture network, with the purpose of making the output of each expert network in the expert mixture network different, thereby improving the accuracy of the predicted value output by the expert mixture network.

[0102] Furthermore, using the encoded feature representation to determine the matching result between the first object and the second object can also include: using a gating network based on a pre-trained type embedding representation to determine the weights corresponding to multiple expert networks; based on the weights corresponding to the multiple expert networks, performing a step of weighted summing of the matching values predicted by the multiple expert networks.

[0103] Continue to refer Figure 3 , Figure 3The resource matching model shown also includes a gating network, which is used to assign corresponding weights to each expert network in the expert mixture network. In the resource matching scenario, the objects under the second type of resource can be further subdivided. For example, if the second type of resource is a job, it can be further subdivided into jobs in different types of industries; for example, if the second type of resource is a house, it can be further subdivided into houses with different types of uses, and so on. Therefore, in order to solve the problem of different standards for judging whether resources are matched under different types, type information for the second type of resource can be introduced, and the type embedding representation corresponding to this type information is trained during the training process of the resource matching model. When determining the matching result between the first object and the second object, the gating network assigns a corresponding weight to each expert network in the expert mixture network based on the type embedding representation, so that the matching values predicted by multiple expert networks are weighted summed according to the corresponding weight of each expert network to obtain the matching value between the first object and the second object, which can improve the accuracy of the expert mixture network in predicting the matching value between the first object and the second object.

[0104] In an embodiment of the present application, sample pairs including positive samples and negative samples can be obtained, wherein the positive samples include sample data of a second object sample and sample data of a first object sample that matches the second object sample, and the negative samples include sample data of the second object sample and sample data of the first object sample that does not match the second object sample. The sample data includes descriptive text and historical interaction sequences.

[0105] The resource matching model is trained using the above training data, wherein the type embedding representation used by the gating network is initialized before training the resource matching model. The embedding representation of the description text of the first object sample and the embedding representation of the historical interaction sequence, and the embedding representation of the description text of the second object sample and the embedding representation of the historical interaction sequence are input into the resource matching model to obtain the matching result of the first object and the second object output by the resource matching model. The loss function is designed using the training objective, and the value of the loss function is used in each round of iteration to update the model parameters and the above type embedding representation using methods such as gradient descent until the preset training end conditions are met. The training end conditions may include, for example, the value of the loss function is less than or equal to a preset loss function threshold, the number of iterations reaches a preset number threshold, etc.

[0106] The training goal is to maximize the matching result between the first object sample and the second object sample in the positive sample, and to minimize the matching result between the first object sample and the second object sample in the negative sample.

[0107] Optionally, the loss function can be a ranking loss function.

[0108] Preferably, the loss function may be a BPR (Bayesian Personalized Ranking) loss function. In addition, a regularization term may be added to the BPR loss function to reduce the overfitting problem of the model.

[0109] In some embodiments, the present application provides a person-job matching method, which can also be applied to Figure 1 The system shown. Figure 5 This is a flow chart of the person-job matching method provided in the embodiment of this application. Figure 5 As shown in , the method may include the following steps:

[0110] Step 501: Obtain the description text of the target candidate, the description text of the target position, the historical interaction sequence of the target candidate, and the historical interaction sequence of the target position. The historical interaction sequence of the target candidate includes at least one position information that has generated a preset type of interaction with the target candidate, and the historical interaction sequence of the target position includes at least one candidate information that has generated a preset type of interaction with the target position.

[0111] Step 503: Encode the description text of the target candidate, the description text of the target position, the historical interaction sequence of the target candidate, and the historical interaction sequence of the target position, and use the feature representation obtained by encoding to determine the matching result between the target candidate and the target position.

[0112] It can be seen from the above process that when this application is applied to person-job matching, in addition to considering the descriptive texts of the target candidate and the target job, it further considers the historical interaction sequence of the target candidate and the historical interaction sequence of the target job, introduces two-way person-job interaction information into person-job matching, and refines the historical interaction sequence through preset type interactions to improve the precision of the feature representation corresponding to the historical interaction sequence, thereby improving the accuracy of person-job matching.

[0113] In one embodiment, the description text of the target candidate is the target candidate resume, and the description text of the target position may also be referred to as the job description information (Job Description, JD for short) of the target position.

[0114] In one embodiment, interaction types can be divided into the following categories based on the degree of match: arranging a resume evaluation, passing a resume evaluation, and passing a full interview. The target candidate's historical interaction sequence then includes: job information for which the target candidate's resume was evaluated, job information for which the target candidate's resume was evaluated and passed, and job information for which the target candidate was interviewed and passed. The target position's historical interaction sequence includes: candidate information for which the resume was evaluated, candidate information for which the resume was passed, and candidate information for which the interview was passed. By fine-grainedly dividing the historical interaction sequence, more accurate feature representations corresponding to the historical interaction sequence can be extracted, thereby further improving the accuracy of determining the match between the target candidate and the target position.

[0115] How to perform encoding in this embodiment and how to use the feature representation obtained by encoding to determine the matching result can be found in the relevant records in the previous embodiments and will not be repeated here.

[0116] It should be noted that the technical solution provided in this application can be applied not only to the person-job matching scenario, but also to the matching scenarios of consumers and commodities, users and participants in activities, suppliers and projects, tenants and houses, beneficiaries and donated resources, students and schools, patients and medical resources, and other resource matching scenarios.

[0117] In some embodiments, the present application also provides a resource matching device. Figure 6 This is a schematic diagram of the structure of a resource matching device provided in an embodiment of the present application. Figure 6 As shown in FIG, the resource matching device 600 may include: an acquisition module 601 and a matching module 602, and may further include an enhancement module 603. The main functions of each component unit are as follows:

[0118] Acquisition module 601 is used to acquire the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object, wherein the first object and the second object belong to the first type of resources and the second type of resources respectively, the historical interaction sequence of the first object includes the object information of at least one second type of resource that generates a preset type of interaction with the first object, and the historical interaction sequence of the second object includes the object information of at least one first type of resource that generates a preset type of interaction with the second object.

[0119] The matching module 602 is used to encode the description text of the first object, the description text of the second object, the historical interaction sequence of the first object and the historical interaction sequence of the second object, and determine the matching result between the first object and the second object using the feature representation obtained by encoding.

[0120] As one possible implementation method, the description text of the second object is the text obtained by enhancing the original description text of the second object by the enhancement module 603 using the large language model.

[0121] The enhancement module 603 may be configured to: if the quality of the original description text of the second object is lower than the preset quality requirement, construct a prompt instruction using the original description text of the second object, the enhancement task information, and the thinking chain prompt template, input the prompt instruction into the large language model, and obtain the enhanced text output by the large language model as the description text of the second object; wherein the thinking chain prompt template includes an inference path for instructing the large language model to perform enhancement.

[0122] As one of the possible implementations, the reasoning path includes:

[0123] Analyze the original description text of the second object to obtain attribute information corresponding to the second object;

[0124] Determine whether the second object has object information of the first type of resource that has been successfully matched historically;

[0125] If so, extract the common information from the object information of the first type of resource that has been successfully matched historically, and use the common information and attribute information to enhance the original description text of the second object to obtain the description text of the second object;

[0126] If it does not exist, the original description text of the second object is enhanced using the attribute information to obtain the description text of the second object.

[0127] As one of the possible implementation methods, when encoding the descriptive text of the first object, the descriptive text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object, the matching module 602 can be specifically configured to: respectively obtain the embedded representation corresponding to the descriptive text of the first object, the embedded representation corresponding to the descriptive text of the second object, the embedded representation corresponding to the historical interaction sequence of the first object, and the embedded representation corresponding to the historical interaction sequence of the second object; use the first encoder to respectively encode the embedded representation corresponding to the descriptive text of the first object and the embedded representation corresponding to the historical interaction sequence of the first object to obtain the feature representation corresponding to the descriptive text of the first object and the feature representation corresponding to the historical interaction sequence of the first object; and use the second encoder to respectively encode the embedded representation corresponding to the descriptive text of the second object and the embedded representation corresponding to the historical interaction sequence of the second object to obtain the feature representation corresponding to the descriptive text of the second object and the feature representation corresponding to the historical interaction sequence of the second object.

[0128] As one possible implementation method, both the first encoder and the second encoder include encoding modules.

[0129] When the first encoder and the second encoder encode the embedded representation corresponding to the description text, the encoding module can be used to encode the embedded representation corresponding to the description text to obtain the feature representation corresponding to the description text.

[0130] As one of the feasible ways, both the first encoder and the second encoder include an encoding module and a fusion module; when encoding the embedded representation corresponding to the historical interaction sequence, the first encoder and the second encoder can use the fusion module to perform multi-head attention processing on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the first object to obtain a first fused representation; perform multi-head attention processing on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the second object to obtain a second fused representation; splice the first fused representation and the second fused representation to obtain a third fused representation; and use the encoding module to encode the third fused representation to obtain a feature representation corresponding to the historical interaction sequence.

[0131] As one possible implementation, the embedded representation is obtained through an embedding network. The training module (not shown in the figure) can be configured as follows:

[0132] Obtain sample pairs including positive samples and negative samples, where the positive samples include sample data of the second object sample and sample data of the first object sample that matches the second object sample, and the negative samples include sample data of the second object sample and sample data of the first object sample that does not match the second object sample, and the sample data include descriptive text and historical interaction sequences; use the general vector model as the initial model of the embedding network, adopt contrastive learning to train the initial model using the sample pairs, and obtain the embedding network.

[0133] As one of the possible implementation methods, when the matching module 602 uses the feature representation obtained by encoding to determine the matching result between the first object and the second object, it can be specifically configured as follows: using multiple expert networks in the expert mixture network to predict the matching value between the first object and the second object based on the feature representation obtained by encoding; and performing weighted summation on the matching values predicted by the multiple expert networks to obtain the matching result between the first object and the second object.

[0134] Furthermore, when the matching module 602 uses the encoded feature representation to determine the matching result between the first object and the second object, it can also be configured to: use the gating network based on the type embedding representation obtained by pre-training to determine the weights corresponding to multiple expert networks; based on the weights corresponding to the multiple expert networks, perform the step of weighted summing of the matching values predicted by the multiple expert networks.

[0135] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or device embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0136] In some embodiments, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of any one of the methods described in the aforementioned method embodiments.

[0137] In some embodiments, the present application also provides an electronic device comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, execute the steps of any of the methods described in the aforementioned method embodiments.

[0138] in, Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 7 As shown, the electronic device 700 may include a processor 710, a video display adapter 711, a disk drive 712, an input / output interface 713, a network interface 714, and a memory 720. The processor 710, the video display adapter 711, the disk drive 712, the input / output interface 713, the network interface 714, and the memory 720 may be communicatively connected via a communication bus 730. The input / output interface 713 may also be referred to as an I / O interface 713.

[0139] Among them, the processor 710 can be implemented by a general-purpose CPU, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., to execute relevant programs to implement the technical solutions provided in this application.

[0140] The memory 720 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 720 can store an operating system 721 for controlling the operation of the electronic device 700, and a basic input and output system (BIOS) 722 for controlling the low-level operations of the electronic device 700. In addition, a web browser 723, a data storage management system 724, and a resource matching device 600, etc. can also be stored. The above-mentioned resource matching device 600 can be an application program that specifically implements the operations of the aforementioned steps in the embodiment of the present application. In short, when the technical solution provided by the present application is implemented by software or firmware, the relevant program code is stored in the memory 720 and is called and executed by the processor 710.

[0141] The input / output interface 713 is used to connect input / output modules to implement information input and output. The input / output modules can be configured as components in the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.

[0142] The network interface 714 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WiFi, Bluetooth, etc.).

[0143] The bus 730 comprises a pathway for transmitting information between the various components of the device (eg, the processor 710 , the video display adapter 711 , the disk drive 712 , the input / output interface 713 , the network interface 714 , and the memory 720 ).

[0144] It should be noted that although the above device only shows the processor 710, video display adapter 711, disk drive 712, input / output interface 713, network interface 714, memory 720, bus 730, etc., in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may also include only the components necessary to implement the solution of the present application, and does not necessarily include all the components shown in the figure.

[0145] In some embodiments, the present application further provides a computer program product, comprising a computer program, which, when executed, implements the steps of any one of the methods described in the aforementioned method embodiments.

[0146] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer program product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0147] The above is a detailed introduction to the technical solutions provided by this application. Specific examples are used herein to illustrate the principles and implementation methods of this application. The description of the above embodiments is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the contents of this specification should not be understood as limiting this application.

Claims

1. A resource matching method, characterized in that: The method comprises: Obtaining a description text of a first object, a description text of a second object, a historical interaction sequence of the first object, and a historical interaction sequence of the second object, wherein the first object and the second object belong to a first type of resource and a second type of resource, respectively, the historical interaction sequence of the first object includes object information of at least one second type of resource that has generated a preset type of interaction with the first object, and the historical interaction sequence of the second object includes object information of at least one first type of resource that has generated a preset type of interaction with the second object; The description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object are encoded, and a matching result between the first object and the second object is determined using a feature representation obtained by encoding.

2. The method according to claim 1, characterized in that The description text of the second object is obtained by enhancing the original description text of the second object using the large language model, wherein enhancing the original description text of the second object using the large language model includes: If the quality of the original description text of the second object is lower than the preset quality requirement, construct a prompt instruction using the original description text of the second object, the enhanced task information, and the thought chain prompt template, input the prompt instruction into the large language model, and obtain the enhanced text output by the large language model as the description text of the second object; The thought chain prompt template includes an inference path for instructing the large language model to perform the enhancement.

3. The method according to claim 2, characterized in that The reasoning path includes: Analyzing the original description text of the second object to obtain attribute information corresponding to the second object; Determine whether the second object has object information of a first-type resource that has been successfully matched historically; If so, extracting common information from the object information of the first type of resource that has been successfully matched historically, and enhancing the original description text of the second object using the common information and the attribute information to obtain a description text of the second object; If not, the original description text of the second object is enhanced using the attribute information to obtain the description text of the second object.

4. The method according to claim 1, wherein Encoding the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object includes: Respectively obtaining an embedded representation corresponding to the description text of the first object, an embedded representation corresponding to the description text of the second object, an embedded representation corresponding to the historical interaction sequence of the first object, and an embedded representation corresponding to the historical interaction sequence of the second object; Using a first encoder, respectively encode the embedded representation corresponding to the description text of the first object and the embedded representation corresponding to the historical interaction sequence of the first object to obtain a feature representation corresponding to the description text of the first object and a feature representation corresponding to the historical interaction sequence of the first object; and A second encoder is used to encode the embedded representation corresponding to the description text of the second object and the embedded representation corresponding to the historical interaction sequence of the second object respectively to obtain a feature representation corresponding to the description text of the second object and a feature representation corresponding to the historical interaction sequence of the second object.

5. The method according to claim 4, characterized in that The first encoder and the second encoder both include an encoding module; Encoding the embedded representation corresponding to the description text includes: The encoding module is used to encode the embedded representation corresponding to the description text to obtain the feature representation corresponding to the description text.

6. The method according to claim 4, characterized in that The first encoder and the second encoder both include an encoding module and a fusion module; Encoding the embedded representation corresponding to the historical interaction sequence includes: Using the fusion module, multi-head attention processing is performed on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the first object to obtain a first fused representation; multi-head attention processing is performed on the embedded representation corresponding to the historical interaction sequence and the embedded representation corresponding to the description text of the second object to obtain a second fused representation; the first fused representation and the second fused representation are concatenated to obtain a third fused representation; The encoding module is used to encode the third fusion representation to obtain a feature representation corresponding to the historical interaction sequence.

7. The method according to claim 4, characterized in that The embedding representation is obtained through an embedding network, which is pre-trained in the following way: Obtaining a sample pair including a positive sample and a negative sample, wherein the positive sample includes sample data of a second object sample and sample data of a first object sample that matches the second object sample, and the negative sample includes sample data of the second object sample and sample data of the first object sample that does not match the second object sample, and the sample data includes a description text and a historical interaction sequence; The general vector model is used as the initial model of the embedding network, and the initial model is trained using the sample pairs in a contrastive learning manner to obtain the embedding network.

8. The method according to any one of claims 1 to 6, characterized in that Determining the matching result between the first object and the second object by using the feature representation obtained by encoding includes: using a plurality of expert networks in a mixture of experts network to predict a matching value between the first object and the second object based on the feature representations obtained by the encoding; The matching values predicted by the multiple expert networks are weighted and summed to obtain a matching result between the first object and the second object.

9. The method according to claim 8, characterized in that The determining the matching result between the first object and the second object using the encoded feature representation further comprises: determining weights corresponding to the plurality of expert networks using a gating network based on a pre-trained type embedding representation; Based on the weights corresponding to the multiple expert networks, a step of weighted summing the matching values predicted by the multiple expert networks is performed.

10. A person-job matching method, characterized in that: include: Obtaining a description text of a target candidate, a description text of a target position, a historical interaction sequence of the target candidate, and a historical interaction sequence of the target position, wherein the historical interaction sequence of the target candidate includes information about at least one position that has generated a preset type of interaction with the target candidate, and the historical interaction sequence of the target position includes information about at least one candidate that has generated a preset type of interaction with the target position; The description text of the target candidate, the description text of the target position, the historical interaction sequence of the target candidate, and the historical interaction sequence of the target position are encoded, and the feature representation obtained by encoding is used to determine the matching result between the target candidate and the target position.

11. A resource matching device, characterized in that: include: an acquisition module, configured to acquire a description text of a first object, a description text of a second object, a historical interaction sequence of the first object, and a historical interaction sequence of the second object, wherein the first object and the second object belong to a first type of resource and a second type of resource, respectively, the historical interaction sequence of the first object includes object information of at least one second type of resource that has generated a preset type of interaction with the first object, and the historical interaction sequence of the second object includes object information of at least one first type of resource that has generated a preset type of interaction with the second object; A matching module is used to encode the description text of the first object, the description text of the second object, the historical interaction sequence of the first object, and the historical interaction sequence of the second object, and use the feature representation obtained by encoding to determine the matching result between the first object and the second object.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the steps of the method according to any one of claims 1 to 10 are implemented.

13. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being configured to store program instructions, wherein when the program instructions are read and executed by the one or more processors, the program instructions execute the steps of the method according to any one of claims 1 to 10.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed, the steps of the method according to any one of claims 1 to 10 are implemented.

Images