Search condition construction method and device, electronic equipment and storage medium

Abstract

The invention relates to a search condition construction method and device, electronic equipment and a storage medium. The method comprises the steps that user input information is acquired; adding the user input information into the prompt information of the large model, and inputting the prompt information into the large model; wherein the large model prompt information comprises a user content information part, a search engine information part and a search condition format information part, and the user input information is added to the user content information part; the search engine information part at least comprises field information used by a search engine during searching; and the large model constructs search conditions conforming to a specified format in the format information part based on user input information in the user content information part according to field information used by the search engine in the prompt information during search, and the search conditions constructed by the large model serve as input information of the search engine. The user intention can be well understood, the information search accuracy is high, and the application range is wide.

Description

Technical Field

[0001] The present invention relates to the field of computer data processing technology, and in particular to a search condition construction method, device, electronic device and storage medium. Background Art

[0002] Information retrieval is a technology that efficiently obtains relevant information from large-scale data. It is widely used in fields such as Internet search, e-commerce, healthcare, knowledge management, and intelligent assistants, and plays a vital role in today's era of information explosion. Current information retrieval systems (or search systems) usually include search engines. The search engine uses a certain search algorithm to retrieve information that meets the requirements from the data source based on the received search requirements. For example, the information retrieval system in an e-commerce platform constructs search conditions based on the product name and product attribute information entered by the user. The search engine matches the products that meet the search conditions in the product database and outputs them to the user. For another example, the information retrieval system in a recruitment platform constructs search conditions based on the job title and job attribute information entered by the job seeker. The search engine then matches the jobs that meet the search conditions in the recruitment job database according to a certain algorithm and outputs them to the job seeker.

[0003] In addition to factors such as the search engine's own search logic and hardware performance, whether the search criteria entered into the search engine accurately and comprehensively understand the user's intent is a key factor affecting the accuracy of the retrieved information. Currently, most information retrieval systems construct search criteria based on user-entered text. With the development of speech-to-text technology, some information retrieval systems construct search criteria based on user voice information. In these information retrieval systems, user voice information is first converted into text. Then, a keyword extraction model is used to extract keywords from the user's voice recognition text. Search criteria information is then generated based on the keywords. For example, Chinese patent application publication number CN104462262B, entitled "A Method, Apparatus, and Browser Client for Implementing Voice Search," discloses a solution that recognizes user-entered voice information to obtain character information that is used as a search term classification. Another example is Chinese patent application publication number CN106959999A, entitled "A Method and Apparatus for Voice Search," which discloses a solution that identifies search keywords from user voice and conducts searches based on the search keywords and search type.

[0004] Existing systems that use keyword extraction models (such as BERT-based named entity recognition models) to extract keywords from user speech recognition text are limited by the model. The extracted keywords cannot adapt to user needs unless the model is retrained. Consequently, this approach fails to accurately capture user intent. Furthermore, the lack of logical connections between the extracted keywords makes it even more difficult to express user intent through a keyword set, especially when the user's input is long and includes multiple keywords. This results in search results that fall short of user expectations in terms of accuracy and relevance. For example, a voice search input from an HR user on a recruitment platform may include requirements at various levels, such as location and skills. Existing solutions can only extract these location and skills keywords and combine them to construct search criteria. However, considering user needs, treating all of these keywords equally for search results is clearly inappropriate. Furthermore, some information is lost during the keyword extraction process, which can lead to misunderstandings of intent. For example, when a user inputs "I don't like people with more than five years of work experience," keyword extraction only extracts "five years of work experience," ignoring the linguistic and logical relationships within the user's context, resulting in inconsistent search results. Summary of the Invention

[0005] In response to the technical problems existing in the prior art, the present invention proposes a search condition construction method, device, electronic device and storage medium to improve the accuracy of understanding user intentions.

[0006] In order to solve the above technical problem, according to one aspect of the present invention, a search condition construction method is provided, which is applied to an information search system, wherein the information search system includes a search engine, and the method includes the following steps:

[0007] Get user input information;

[0008] Adding user input information to the large model prompt information and inputting the prompt information into the large model; wherein the large model prompt information includes a user content information portion, a search engine information portion, and a search condition format information portion, the user input information is added to the user content information portion; the search engine information portion includes at least field information used by the search engine during the search; and

[0009] The large model constructs search conditions that conform to the format specified in the format information section based on the field information used by the search engine in the prompt information and the user input information in the user content information section, wherein the search conditions constructed by the large model serve as input information for the search engine.

[0010] Optionally, when the user input information is voice information, the method further includes: converting the voice information into text information; correspondingly, adding the text information converted from the voice information to the large model prompt information.

[0011] Optionally, when converting the voice information into text information, a voice-to-text service interface provided by a third-party system is called, and the user's voice information is sent to the third-party system through the voice-to-text service interface, and text information returned by the third-party system is received.

[0012] Optionally, the prompt information of the large model further includes an example portion, wherein the example portion includes one or more example information, each example information including user input information and a search condition that complies with a format specified in the format information portion.

[0013] Optionally, the example portion of the large model prompt information includes multiple scene example sub-portions, and each scene example sub-portion includes one or more example information.

[0014] Optionally, the search condition construction method further includes:

[0015] Determine the user's current scenario based on user input information;

[0016] Correspondingly, the scene example subsection corresponding to the current scene is marked in the example section in the large model prompt information to prompt the large model to construct a search condition according to the marked scene.

[0017] Optionally, the search engine information part further includes identification information of the search engine and business meaning description information of the field.

[0018] Optionally, the search engine information section includes a plurality of subsections, each subsection including identification information of a search engine, field information used by the search engine, and business meaning description information thereof.

[0019] Optionally, the search condition construction method further includes:

[0020] Obtaining search engine information of the information search system;

[0021] The search engine information of the information search system is based on the search engine information, and the identification information of the search engine is marked to prompt the large model to construct the search condition according to the field information of the marked search engine.

[0022] Optionally, the large model prompt information also includes a field supplement section, which includes field supplementary description information; correspondingly, when the large model constructs search conditions that conform to the specified format in the format information section based on the user input information in the user content information section according to the field information used by the search engine in the prompt information, when the large model cannot match the corresponding field information for the user input information, supplementary field information is generated based on the input information.

[0023] Optionally, after generating the supplementary field information, the large model further includes: saving the supplementary field information and the corresponding user information as information to be verified; or, building the supplementary field information and the corresponding user information into the search conditions, and marking the supplementary field information.

[0024] Optionally, the search condition construction method further includes:

[0025] Calling a large model interface provided by a third party, inputting the large model prompt information into the large model provided by the third party via the large model interface, and receiving the search conditions constructed by the large model via the large model interface; or,

[0026] The large model is deployed locally in the information search system, and the large model prompt information is input into the locally deployed large model; and the search conditions output by the locally deployed large model are received.

[0027] Optionally, the search conditions include fields and field values ​​expressed in a specified format, the fields are fields used by the search engine when searching, and the field values ​​are information extracted from user input information.

[0028] According to another aspect of the present invention, the present invention also provides a search condition construction device, which is applied to an information search system, wherein the information search system includes a search engine, and the device includes a user input information module, a prompt information generation module and a large model module, wherein the user input information module is configured to obtain user input information; the prompt information generation module is configured to add the user input information to the large model prompt information; wherein the large model prompt information includes a user content information part, a search engine information part and a format information part of the search condition, and the user input information is added to the user content information part; the search engine information part at least includes field information used by the search engine when searching; the large model module is configured to input the prompt information into the large model, and the large model constructs search conditions that conform to the specified format in the format information part based on the user input information in the user content information part according to the field information used by the search engine in the prompt information; wherein the search conditions constructed by the large model serve as input information of the search engine.

[0029] According to another aspect of the present invention, the present invention also provides an electronic device, which includes a processor and a memory, wherein a computer program instruction set is stored on the memory, and the aforementioned search condition construction method is implemented when the processor executes the computer program instruction set on the memory.

[0030] According to another aspect of the present invention, the present invention further provides a computer-readable storage medium, wherein a computer program instruction set is stored on the computer-readable storage medium, and when the computer program instruction set is executed by a processor, the aforementioned search condition construction method is implemented.

[0031] According to another aspect of the present invention, the present invention further provides a computer program product, which includes a computer program instruction set, and when the computer program instruction set is executed by a processor, the above-mentioned search condition construction method is implemented.

[0032] The present invention can better understand user intentions, has high information search accuracy, can effectively improve user search experience, and has a wide range of applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Below, the preferred embodiments of the present invention will be further described in detail with reference to the accompanying drawings, in which:

[0034] Figure 1 is a principle block diagram of an information search system according to an embodiment of the present invention;

[0035] Figure 2 is a flow chart of a method for constructing search conditions according to one embodiment of the present invention;

[0036] Figure 3 is a flow chart of a resume search method according to one embodiment of the present invention;

[0037] Figure 4 is a flow chart of a method for constructing search conditions according to another embodiment of the present invention;

[0038] Figure 5 is a principle block diagram of a search condition construction device according to an embodiment of the present invention;

[0039] Figure 6 is a principle block diagram of a search condition construction device according to another embodiment of the present invention; and

[0040] Figure 7 FIG. 1 is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0042] In the detailed description that follows, reference may be made to the various drawings that form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Each specific embodiment of the present application is described below in sufficient detail to enable a person of ordinary skill in the art to implement the technical solutions of the present application. It should be understood that other embodiments may be utilized or that structural, logical, or electrical changes may be made to the embodiments of the present application.

[0043] Information retrieval is a necessary means of obtaining information in the information explosion era. In this invention, the systems used for information retrieval in various platforms are collectively referred to as information search systems. Figure 1 , Figure 1 This is a block diagram of the principles of an information search system according to an embodiment of the present invention. In this embodiment, the information search system 100 includes a user end and a server end. The user end is located in a user terminal device 101, which includes a user interaction module. The user terminal device 101 is such as a smart phone, a desktop computer, a laptop computer, etc. The server end is installed in a server 102, which includes one or more search engines. The user inputs information, such as a text or a voice, through the user interaction module of the user end. The user end sends the information input by the user to the server end. The server end constructs search conditions based on the user input information. The search engine uses the corresponding algorithm based on the search conditions to retrieve information that meets the user's needs in the database, and then returns the information to the user end and displays it on the current interface. Figure 1 The information search system in the example is an information search system in a recruitment platform, and the information searched is, for example, job information and resume information. The information search system can also be an information search system in an e-commerce platform, and the information searched is, for example, product information.

[0044] Figure 2 1 is a flow chart of a method for constructing search conditions according to an embodiment of the present invention. The method comprises the following steps:

[0045] Step S11, obtaining user input information.

[0046] Step S12: Add the user input information to the large model prompt information and input the prompt information to the large model.

[0047] Step S13: receiving the search conditions constructed by the large model and sending them to the search engine as input information of the search engine.

[0048] In this embodiment, the user input information may be, for example, a text message input by the user, or a voice message input by the user. When the user input information is a voice message, the voice message is first converted into text information, and then in step S12, the text information converted from the voice message is added to the large model prompt information.

[0049] When converting the voice information into text information, in one embodiment, a traditional speech recognition method is used, for example, a statistical method using a Gaussian mixture model (GMM) and a hidden Markov model (HMM) is used to convert the voice information into text information. Alternatively, the voice information can be converted into text information based on a deep learning framework, for example, a hybrid model based on DNN-HMM, an end-to-end deep learning model, such as a CNN / RNN / Transformer acoustic model, CTC (Connectionist Temporal Classification), an enhanced language model (LM), such as a neural language model (BERT / GPT) and a large model, such as Whisper, Gemini, etc. In another embodiment, when converting the voice information into text information, a voice-to-text service interface provided by a third-party system is called, and through the voice-to-text service interface, the user's voice information is sent to the third-party system, and the text information returned by the third-party system is received. In this embodiment, the voice-to-text service interface is, for example, a voice-to-text API provided by the Baidu AI open platform. This embodiment fully utilizes the advantages of the voice-to-text service provided by the third-party system and reduces the processing pressure of the system itself.

[0050] The macro model prompt information includes a user content information section, a search engine information section, and a search condition format information section. In step S12, the user input information is added to the user content information section. The search engine information section includes at least field information used by the search engine during searches. After the prompt information is input to the macro model, the macro model constructs search conditions that conform to the format specified in the format information section based on the user input information in the user content information section and the field information used by the search engine during searches in the prompt information.

[0051] The large model in the present invention is also commonly referred to as a large language model (LLM), such as the Generative Pre-Trained Transformer (GPT) series model developed by Open AI, LaMDA, PaLM developed by Google, LLaMA developed by Meta, Wenxin Yiyan developed by Baidu, Tongyi Qianwen of Alibaba, Hunyuan of Tencent, the large language model named "AliceMind" developed by Alibaba Damo Academy, the deepseek model (deepseek-ai / DeepSeek-R1-Distill-Qwen-14B) developed by Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., etc. Developers in the field of large models can apply any LLM according to actual conditions. By training on large-scale text data, the large model can learn knowledge such as grammar and semantics of the language. When applied, the large model can analyze the complete semantics of the sentence, not just extract keywords. In the present invention, the large model can mine the user's complex needs, such as time range, logical relationships, and other information hidden in the text.

[0052] In the present invention, when applying a large model, in one embodiment, a large model interface provided by a third party, such as the interfaces provided by the various AI platforms mentioned above, is called. In step S12, the large model prompt information is input to the large model provided by the third party via the large model interface, and the search conditions constructed by the large model are received via the large model interface. The search conditions are then sent to the search engine. In another embodiment, the large model is locally deployed in the information search system, and the large model prompt information is input to the locally deployed large model. After the large model generates the search conditions, it sends the constructed search conditions to the search engine.

[0053] In a further embodiment, in order to enable the large model to better understand the instructions to generate search conditions that meet the requirements, the prompt information also includes an example part, which includes one or more example information, each example information includes user input information and search conditions that meet the format specified in the format information part.

[0054] To enable the information search system to provide users with accurate search information in different usage scenarios, in a further embodiment, the example section of the macro model prompt information includes multiple scenario example subsections, each of which includes one or more example information. For example, on a recruitment platform, this includes a job search scenario on the job seeker's homepage, a resume search scenario on the recruiter's homepage, and search scenarios within the user's homepage. Another example includes a product search scenario on an e-commerce platform, a product search scenario within a specific product category page, and a search scenario within the user's order page. This embodiment of the present invention explicitly describes the current scenario and its characteristics in the example information corresponding to the scenario. For example, in a resume search scenario, the example information in the prompt information may read: "The current scenario is a resume search scenario. HR will enter a search term. This search term may include multiple requirements, such as educational qualifications (corresponding to the search field "top degree") or age (corresponding to the search field "age")..." This allows the macro model to generate accurate search criteria based on the specific user input information based on this example. When the macro model prompt information includes multiple scenario example information, it can determine whether the current scenario is a job search scenario or a resume search scenario based on whether the user input information is searching for a position or a resume. In addition, the information search system may include a front-end (including an interactive module on the user side) and a back-end (i.e., a server side). The execution body of the search condition construction method of the present invention is the back-end, and the search request received by the back-end from the front-end includes user input information and scenario information. For example, when a user enters a message "xxxx" in the search entry on the user side, the front-end will send a request string to the back-end server. The content in the request string is, for example, {"user search content":"xxxx","current scenario":"search position"}. After the back-end receives this search request, it determines the specific scenario from the content corresponding to the field "current scenario" in the request string, such as if the scenario corresponding to the request is "search position".

[0055] After determining the specific scenario, the scenario example subsection corresponding to the current scenario is annotated in the example section of the macro model prompt information to prompt the macro model to construct search conditions based on the annotated scenario. Optionally, after determining the specific scenario, example information corresponding to other scenarios in the example subsection of the prompt information can be deleted, retaining only the example information corresponding to the current scenario.

[0056] In another embodiment, the current scene is sent to the big model as a type of information in the user content information part. The big model determines the scene example information for reference based on the current scene, and then constructs the search conditions based on the example information.

[0057] In order to reduce the workload of the search engine to reorganize the search conditions, the present invention can generate search conditions that conform to a specific format and can be directly executed by the search engine, and send them directly to the search engine as input information. Specifically, the large model prompt information includes a format information part of the search conditions, in which the format is specified, so that the large model can output the search condition information in the specified format when constructing the search conditions. For example, when the ES search engine (Elasticsearch) is used, the corresponding specified format is DSL (Domain Specific Language). DSL is a programming language designed specifically for a specific domain or application. The syntax and semantics of DSL are designed around specific tasks (such as the ES search engine in this embodiment for search recall), so it is more efficient and accurate in expressing search conditions, and can accurately convert the user intention understood by the large language model into instructions that can be executed by the search engine.

[0058] For example, when building a search engine using Apache Solr, the format specified by the format information section conforms to Solr query statements. For example, when HR searches for "Looking for talent in automotive seat hardware development," the corresponding Solr query statement is: q = (industry: "automobile" AND job_function: "seat hardware development").

[0059] Corresponding to the different search information, the search engine has a corresponding search field information library, which includes fields used in the search. For example, the search field information library of the search engine in the recruitment platform has fields such as "industry", "job title", "location", "salary", etc., and the search field information library of the search engine in the e-commerce platform has fields such as "product category", "product name", "price range", etc.

[0060] The present invention adds the field information in the search field information library of the search engine application to the search engine information part in the large model prompt information, so that the large language model converts the user intention it understands into instructions that can be executed by the search engine. These instructions include the search fields of the search engine application and their field values.

[0061] In addition, in order to enable the big model to accurately understand the meaning of the search field of the search engine application, in one embodiment, the search engine information part of the big model prompt information also includes description information of the field and its business meaning, so that the big model can better understand the meaning of the field when parsing user input information and matching the search field.

[0062] Furthermore, the search engine information section of the large model prompt information also includes search engine identification information, which corresponds to the fields used by the search engine and the business meaning of the fields. The identification information may include, for example, the search engine name, the search framework name used by the search engine, etc.

[0063] When the information search system includes multiple search engines, the search engine information section includes multiple subsections, each subsection including identification information of a search engine, field information used by the search engine, and business meaning description information thereof.

[0064] In one embodiment, after obtaining the user input information in step S11, the user's current scenario and the search engine information of the information search system are also obtained. The identification information of the search engine of the search engine information section is annotated according to the search engine information used by the information search system, so as to prompt the big model to construct the search conditions according to the field information of the annotated search engine. In another embodiment, the current scenario and the search engine information of the information search system (such as the search engine identification) are delivered to the big model together as another type of information in the user content information section, and the big model determines the search engine to be used based on the current scenario, and then understands the search logic of the search engine based on the field information, business meaning description information and one or more example information in the corresponding scenario example subsection in the search engine information section, thereby determining a set of fields that meet the user's search intention and their field values ​​from the user input information based on the understanding of the user input content, thereby constructing the search conditions.

[0065] In traditional search methods, users can only influence the sorting results based on the existing search filter items provided by the front-end. However, the user's needs are sometimes not met by the existing filter items, and the filter requirements of different users vary greatly. The present invention uses a large model to understand the user's search needs and converts the filter requirements of different users with different focuses into filter conditions for the corresponding fields in the search engine. For example, when an HR needs to find talents in the automotive industry who are engaged in the hardware development of car seats, and the input information is "Please find talents who are engaged in the hardware development of car seats", the DSL statement of the search condition constructed by the large model is as follows:

[0066]

[0067]

[0068] In another scenario where HR needs to find talent with three or more years of real estate sales experience, when the input is "Please find talent with three or more years of real estate sales experience," the DSL statement of the search condition constructed by the big model is as follows:

[0069]

[0070]

[0071] The large model can determine appropriate search fields based on user input and construct logically clear and comprehensive search criteria. Traditional keyword extraction methods only use extracted keywords to determine corresponding field values ​​and cannot determine the logical relationships between fields. Therefore, the search criteria constructed by this invention can better express the user's search intent.

[0072] Figure 3 The following is a flow chart of a resume search method according to one embodiment of the present invention. In this embodiment, the information search system is a system used within a recruitment platform to provide resume search services to HR personnel. The search engine used is Elasticsearch, a distributed, RESTful search engine based on the Lucene library that can rapidly store, search, and analyze large amounts of data. Elasticsearch uses DSL statements to describe search criteria, searches massive amounts of data based on those criteria, and uses efficient inverted indexing and other techniques to quickly find data matching the search criteria. In this embodiment, Elasticsearch is used as the search engine to retrieve resumes that meet the criteria.

[0073] Through the user terminal, the user can enter "Find cost accountants working in the medical device industry in Shanghai who have been active in the past week" by voice on the search page of the recruitment platform.

[0074] After the recruitment platform's information search system receives the search voice message, it will process the following steps:

[0075] Step S21 calls the speech-to-text service interface to convert the search voice information into text information. For example, the converted text information is: "Search for cost accountants working in the medical device industry in the Shanghai area who were active in the past week." In this embodiment, the speech-to-text API provided by the Baidu AI open platform is called. In a further embodiment, the converted text information can also be further processed, for example: removing colloquial affixes in the text content (such as "this" and "that" often used by the speaker when they cannot think of appropriate expressions during the expression process), eliminating repeated content (such as content repeated by the speaker for coherence), unifying the language of the text content (such as when the speaker inserts English words in Chinese, causing the interview text to include both Chinese and English. During the unified processing, the minority language content is translated into the majority language content based on the context), etc. According to this embodiment, even if the voice input is of low quality (incoherent language expression, colloquialism, mixing of different language systems, etc.), it is still possible to obtain an accurate text of the user's expression.

[0076] Step S22, add the user input information to the large model prompt information. One of the purposes of the large model prompt information is to enable the large language model to fully understand the user's intention, and the second purpose is to be able to generate correct DSL query statements that can be executed. For example: the large model prompt information includes the task description "You are now an intelligent resume search assistant on the recruitment platform. You need to combine the user's search content and the fields currently available in our search engine to generate a DSL statement that meets the user's search needs and can be correctly executed." The large model prompt information also includes a user content information section, such as "The following is the user's search content: [user voice-to-text content]." The model prompt information also includes a search engine information section, such as "The following is search engine information [engine information identifier, field-business meaning description information]." Among them, the "DSL statement" in the last sentence of the task description is in a specified format, that is, the format information section of the search conditions mentioned above in the present invention. Of course, this information can also be described in a separate paragraph, such as "output DSL statement."

[0077] Step S23: Call the large model interface, send the large model prompt information to the large model, and receive the returned DSL statement expressing the search condition, as follows:

[0078]

[0079]

[0080] Among them, "query," "bool," "must," "match," and "range" are DSL grammatical terms, "industry," "jobTitle," "location," "gte," and "lastActiveDate" are field information used by the search engine during searches, and "medical equipment," "cost accounting," "Shanghai," and "now-1w" are the field values ​​corresponding to the respective fields determined from the user input information. This embodiment uses the syntax of DSL statements to express query methods for different fields. For example, the aforementioned "bool," "must," "match," and "range" correspond to different query methods. Thus, compared to keyword query methods in the prior art, the present invention not only provides the search engine with keywords in the user input information, but also provides query methods corresponding to different keywords (i.e., search field values ​​in this embodiment). In addition, the syntax of DSL query statements also supports compound queries, i.e., multiple leaf queries (querying specific values ​​of specific fields) are combined to form corresponding logical relationships through "and," "or," and other operations. Therefore, the query statement can express the logical relationship between some key information in the user input information, thereby achieving the purpose of accurately expressing the user's intent.

[0081] Step S24: Send the search conditions described by the DSL sentences generated by the large language model to the search engine.

[0082] In step S25, the search engine executes the DSL statement to obtain the first resume set. Specifically, the search engine uses the information in the "lastActiveDate" field as the query scope to query the resume data index for resumes that were active in the past week. It then searches for matching resumes based on the values ​​of fields such as "industry," "jobTitle," and "location," thereby obtaining a resume set that meets the user's search intent.

[0083] In step S26, a preset number of resumes are determined from the first resume set according to a certain algorithm or logical rules as search results. For example, a portion of resume information may be screened out based on user preferences, or based on resume popularity, etc. Gradient boosting tree (GBDT / XGBoost / LightGBM), logistic regression (LR) models, deep learning models, multi-task learning models, etc. may be further used to perform refined sorting based on fine features, ultimately determining a suitable number of resumes as search results.

[0084] Step S27: Return the search results to the user terminal to present them to the user.

[0085] Figure 4FIG. 1 is a flow chart of a method for constructing search conditions according to another embodiment of the present invention. The method comprises the following steps:

[0086] Step S31, obtaining user input information.

[0087] Step S32: Add the user input information to the macro model prompt information and input the prompt information to the macro model. In this embodiment, the macro model prompt information also includes a field supplement section, which includes field supplementary description information. For example, the field supplement section is included in the task description of the prompt information to prompt the macro model to generate corresponding fields when matching the user input information with search engine field information and finding no suitable search engine fields.

[0088] Step S33: The large model constructs search conditions. Specifically, the large model constructs search conditions that conform to the format specified in the format information section based on the field information used by the search engine in the prompt information and the user input information in the user content information section.

[0089] Step S34, determining whether the return information of the large model contains supplementary field information. If so, executing step S35; if not, executing step S36.

[0090] In step S35, the supplementary field information and the corresponding user information are saved as information to be verified, and step S36 is executed.

[0091] Step S36: Send the search conditions in the returned information to the search engine.

[0092] Among them, the large model can return the supplementary field information in the form of a separate file, or it can return the supplementary field information together with the search condition information, and isolate it from the search condition information in a segmented manner. In this embodiment, after receiving the supplementary field information returned by the large model, it is saved as information to be verified so that the developer can verify whether the supplementary field is suitable as a search field for the search engine. In another embodiment, after the large model generates the supplementary field information, the supplementary field information and the corresponding user information are constructed into the search conditions, and the supplementary field information is marked. When the search engine searches based on the search conditions, it searches based on all the current fields and their field values ​​and the logic between the fields to obtain an information set, and at the same time adds the marked supplementary fields to the field information library. This embodiment can not only construct search conditions for users, but also expand the fields of the search engine, thereby improving the search capability of the search engine.

[0093] On the other hand, the present invention also provides a search condition construction device, see Figure 5 , Figure 5This is a principle block diagram of a search condition construction device according to an embodiment of the present invention, which is applied to an information search system, wherein the information search system includes a search engine, and the device includes a user input information module 11, a prompt information generation module 12 and a large model module 13. The user input information module 11 is used to obtain user input information. When the user input information is voice information, the user input information module 11 converts the voice information into text information, for example, by calling a voice-to-text service interface provided by a third-party system, and sending the user's voice information to the third-party system through the voice-to-text service interface, and receiving the text information returned by the third-party system. Alternatively, a voice recognition module is used to convert the voice information into text information. The voice recognition module is a module that uses a traditional voice recognition algorithm, such as a statistical method using a Gaussian mixture model (GMM) and a hidden Markov model (HMM) to convert voice information into text information, or it can be various voice recognition modules constructed using a deep learning framework.

[0094] The prompt information generation module 12 is used to generate large model prompt information and add user input information to the large model prompt information. The large model prompt information includes a user content information section, a search engine information section, and a search condition format information section. The user input information is added to the user content information section. The search engine information section includes at least field information used by the search engine during searches, that is, field information from the search field information library used by the search engine is added to the search engine information section. In one embodiment, after the large model prompt information is generated, it is saved as a template. Each time it is applied, only the current user input information is added to the user content information section or some corresponding adjustments are made.

[0095] The big model module 13 is used to input the prompt information into the big model, and the big model constructs the search conditions that conform to the format specified in the format information part based on the field information used by the search engine in the prompt information and the user input information in the user content information part. The search conditions constructed by the big model serve as the input information of the search engine. In one embodiment, the big model module 13 calls the big model interface provided by a third party, inputs the big model prompt information into the big model provided by the third party through the big model interface, and receives the search conditions constructed by the big model through the big model interface; in another embodiment, the big model module 13 includes a big model locally deployed in the information search system, and after receiving the prompt information sent by the prompt information generation module 12, inputs the big model prompt information into the locally deployed big model, receives the search conditions output by the locally deployed big model, and then outputs the search conditions to the search engine.

[0096] When there are multiple different search engines, the prompt information generation module 12 adds the search engine identifier of the current application in the search engine information section, and marks the format information corresponding to the search engine identifier in the format information section of the search condition. Alternatively, when generating the large model prompt information, the prompt information generation module 12 includes the corresponding information of all search engines used in the information search system. For example, the search engine information section is set to multiple subsections, each subsection includes the identification information of a search engine, the field information used by the search engine, and its business meaning description information. Correspondingly, each format information in the format information section of the search condition is marked with the identification information of the search engine corresponding to it, and the corresponding scenario and / or search engine identification information is marked in the example section. This improves the versatility of the large model prompt information, and it is not necessary to generate the large model prompt information from scratch each time the search condition is constructed. Instead, corresponding markings can be made in the large model prompt information according to the specific circumstances of the current scenario.

[0097] See also Figure 6 , Figure 6 This is a block diagram of the principles of a search condition construction device according to another embodiment of the present invention. In this embodiment, when generating the large model prompt information, the prompt information generation module 12 also adds a field supplement section, which includes field supplementary information. This section is used to prompt the large model to generate corresponding fields when matching the user input information with the search engine field information. Therefore, when the large model generates supplementary fields for the user input information that cannot match the existing fields during the search condition construction process, the generated supplementary fields are provided in a file or in a section with the search condition information. As needed, these supplementary fields are stored in a designated location for developers to review and decide whether to add them to the search field information library.

[0098] The search condition construction method and system provided by the present invention correspond to the current diversification of information search methods and search technologies, and can be well adapted to the user's voice input method. The powerful semantic understanding ability of the general large model can better understand the user's intentions, and bring users a more efficient and accurate search experience. When users conduct information searches, for example, when recruiting users on a recruitment platform search for resumes, recruiting users (such as HR of a recruiting unit) can use voice and complex expressions to describe the resume requirements that match the recruited positions, which not only improves the efficiency of information input, but also does not need to worry about the complexity of the information. The present invention utilizes the powerful semantic understanding ability of the large model, combined with the unique large model prompt information of the present invention, to enable the large model to accurately understand the user's search intentions, match the search engine's field values ​​according to the user's search intentions and needs, and express the logical associations between fields based on the user's search intentions, without causing the problem of the keyword extraction model in the prior art losing information when extracting keywords. Therefore, the constructed search conditions can accurately express the user's search intentions, thereby improving the accuracy of searching resumes.

[0099] In another aspect, the present invention further provides an electronic device, see Figure 7 , Figure 7 This is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present invention. The electronic device can be implemented as a server or various other terminal devices, such as a desktop personal computer, a tablet computer, a laptop computer, a mobile phone, etc., and includes a processor 601 and a memory 602. The memory 602 stores a program instruction set, and when the processor 601 executes the program instruction set on the memory 602, any of the aforementioned search condition construction methods is implemented.

[0100] Specifically, the processor 601 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiment of the present invention.

[0101] The memory 602 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 602 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 602 may include removable or non-removable (or fixed) media. Where appropriate, the memory 602 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 602 is a non-volatile solid-state memory.

[0102] The memory may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Thus, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it implements any of the aforementioned search condition construction methods.

[0103] In one example, the electronic device may further include a communication interface 603 and a bus 604. The processor 601, the memory 602, and the communication interface 603 are connected via the bus 604 and communicate with each other.

[0104] The communication interface 603 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiment of the present invention.

[0105] Bus 604 includes hardware, software or both, and the components of online data flow metering equipment are coupled to each other. For example, but not limitation, bus can include accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnect (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 604 can include one or more buses. Although the embodiment of the present invention describes and shows a specific bus, the present invention considers any suitable bus or interconnection.

[0106] The present invention also provides a computer-readable storage medium having computer program instructions stored thereon, and the computer program instructions can be executed by a processor to implement any one of the search condition construction methods in the aforementioned embodiments. The computer-readable storage medium can be any medium that can tangibly contain or store computer-executable instructions for use by or in combination with an instruction execution system, device, and apparatus. The storage medium can be a transient computer-readable storage medium or a non-transient computer-readable storage medium. Non-transient computer-readable storage media may include, but are not limited to, magnetic storage devices, optical storage devices, and / or semiconductor storage devices. Examples of such storage devices include, for example, magnetic disks, optical disks based on CD, DVD, or Blu-ray technology, and persistent solid-state memories such as flash memory, solid-state drives, and the like.

[0107] The present invention also provides a computer program product comprising a set of computer program instructions that, when executed by a processor, implement any of the search condition construction methods described in the aforementioned embodiments. The computer program product includes, but is not limited to, an application installation package published on a website or in an app store, an application plug-in, or a mini-program that can be run within certain applications.

[0108] The above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the scope of the present invention. Therefore, all equivalent technical solutions should also fall within the scope of the present invention.

Claims

1. A search condition construction method, characterized in that: The method is applied to an information search system, wherein the information search system includes a search engine, and the method includes: Get user input information; Adding user input information to the large model prompt information and inputting the prompt information into the large model; wherein the large model prompt information includes a user content information portion, a search engine information portion, and a search condition format information portion, the user input information is added to the user content information portion; the search engine information portion includes at least field information used by the search engine during the search; and The large model constructs search conditions that conform to the format specified in the format information section based on the field information used by the search engine in the prompt information and the user input information in the user content information section, wherein the search conditions constructed by the large model serve as input information for the search engine.

2. The search condition construction method according to claim 1, characterized in that: When the user input information is voice information, the method further includes: converting the voice information into text information; correspondingly, adding the text information converted from the voice information into the large model prompt information.

3. The search condition construction method according to claim 2, characterized in that: When converting the voice information into text information, a voice-to-text service interface provided by a third-party system is called, and the user's voice information is sent to the third-party system through the voice-to-text service interface, and text information returned by the third-party system is received.

4. The search condition construction method according to claim 1, wherein: The prompt information of the large model further includes an example portion, wherein the example portion includes one or more example information, each example information including user input information and a search condition that complies with a format specified in the format information portion.

5. The search condition construction method according to claim 4, characterized in that: The example portion of the large model prompt information includes multiple scene example sub-portions, and each scene example sub-portion includes one or more example information.

6. The search condition construction method according to claim 5, characterized in that: Further including: Determine the user's current scenario based on user input information; Correspondingly, the scene example subsection corresponding to the current scene is marked in the example section in the large model prompt information to prompt the large model to construct a search condition according to the marked scene.

7. The search condition construction method according to claim 1, characterized in that: The search engine information part also includes the identification information of the search engine and the business meaning description information of the field.

8. The search condition construction method according to claim 7, characterized in that: The search engine information section includes a plurality of subsections, each of which includes identification information of a search engine, field information used by the search engine, and business meaning description information thereof.

9. The search condition construction method according to claim 7 or 8, characterized in that: Further including: Obtaining search engine information of the information search system; The search engine information of the information search system is based on the search engine information, and the identification information of the search engine is marked to prompt the large model to construct the search condition according to the field information of the marked search engine.

10. The search condition construction method according to claim 1, wherein: The large model prompt information also includes a field supplement section, which includes field supplementary description information; correspondingly, when the large model constructs search conditions that conform to the specified format in the format information section based on the user input information in the user content information section according to the field information used by the search engine in the prompt information, if the user input information cannot be matched to the corresponding field information, supplementary field information is generated based on the input information.

11. The search condition construction method according to claim 10, characterized in that: After generating the supplementary field information, the large model further includes: saving the supplementary field information and the corresponding user information as information to be verified; or, building the supplementary field information and the corresponding user information into the search conditions, and marking the supplementary field information.

12. The search condition construction method according to claim 1, wherein: Further including: Calling a large model interface provided by a third party, inputting the large model prompt information into the large model provided by the third party via the large model interface, and receiving search conditions constructed by the large model via the large model interface; or, The large model is deployed locally in the information search system, and the large model prompt information is input into the locally deployed large model; and the search conditions output by the locally deployed large model are received.

13. The search condition construction method according to claim 1, wherein: The search conditions include fields and field values ​​expressed in a specified format. The fields are fields used by the search engine when searching, and the field values ​​are information extracted from user input information.

14. A search condition construction device, characterized in that: The device is applied to an information search system, wherein the information search system includes a search engine, and the device includes: A user input information module configured to obtain user input information; a prompt information generation module configured to add user input information to the large model prompt information; wherein the large model prompt information includes a user content information portion, a search engine information portion, and a search condition format information portion, the user input information being added to the user content information portion; the search engine information portion at least including field information used by the search engine during search; and The big model module is configured to input prompt information into the big model, and the big model constructs search conditions that conform to the specified format in the format information part based on the field information used by the search engine in the prompt information and the user input information in the user content information part; wherein the search conditions constructed by the big model serve as input information of the search engine.

15. An electronic device comprising a processor and a memory, wherein the memory stores a computer program instruction set, characterized in that: The search condition construction method according to any one of claims 1 to 13 is implemented when the processor executes the computer program instruction set on the memory.

16. A computer-readable storage medium, wherein: The computer-readable storage medium stores a computer program instruction set, wherein the computer program instruction set, when executed by a processor, implements the search condition construction method according to any one of claims 1 to 13.

17. A computer program product comprising a set of computer program instructions, characterized in that When the computer program instruction set is executed by a processor, the search condition construction method according to any one of claims 1 to 13 is implemented.

Images