Method and device for screening objects and electronic equipment
By combining a lightweight semantic vectorization model and a large language model, the problem of relying on manual judgment in the object screening process in existing technologies is solved, achieving efficient and accurate object screening and improving the objectivity and consistency of the screening results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-17
- Publication Date
- 2026-04-07
AI Technical Summary
In existing technologies, the process of screening objects relies heavily on human intervention and experience-based judgment, resulting in highly subjective, inefficient, and inaccurate screening results, making it difficult to achieve efficient and precise object screening.
By combining a lightweight semantic vectorization model and a large language model, semantic vectorization matching and in-depth analysis of candidate objects and requirement parameters are achieved, replacing subjective human judgment.
It improves the consistency, stability and efficiency of the screening process, significantly enhances the objectivity and accuracy of the matching results, and realizes fully automated decision-making from information acquisition to in-depth evaluation.
Smart Images
Figure CN121807925A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of Internet technology, and in particular to a method, apparatus, and electronic device for filtering objects. Background Technology
[0002] In business practice, it is often necessary to filter out objects that meet specific criteria from a large candidate set for subsequent decision-making or operations. Existing technologies mainly adopt a screening model dominated by human experience. Typically, based on the user's requirements, the candidate set is generated by matching in a database according to pre-defined rules. Finally, domain experts, combining their subjective experience, manually review, adjust, and sort the candidate set to determine the final target object.
[0003] However, the entire screening process of existing technologies relies heavily on human intervention and experience-based judgment, resulting in highly subjective and inconsistent screening results, making it difficult to achieve efficient and accurate object screening. Summary of the Invention
[0004] This application provides a method, apparatus, and electronic device for screening objects, in order to solve the problems of high subjectivity, low efficiency, and low accuracy in screening results caused by reliance on human intervention and experience judgment in the prior art.
[0005] In a first aspect, embodiments of this application provide a method for filtering objects, including:
[0006] Based on the received user demand parameters, at least one candidate object and the feature information corresponding to each candidate object are obtained from the information database;
[0007] For any candidate object, after determining the first matching parameter based on the requirement parameter and the feature information corresponding to the candidate object, the requirement parameter, the feature information and the first matching parameter are input into the large language model to obtain the second matching parameter corresponding to the candidate object. The first matching parameter represents the similarity between the feature information and the requirement parameter, and the second matching parameter represents the degree to which the candidate object meets the requirement parameter.
[0008] Based on the second matching parameter corresponding to each candidate object, a target object that meets the preset filtering conditions is determined from the at least one candidate object.
[0009] In one possible implementation, the feature information includes structured feature information of the candidate object;
[0010] The step of determining the first matching parameter of any candidate object based on the requirement parameters and the feature information corresponding to any candidate object includes:
[0011] Based on a lightweight semantic vectorization model, the requirement parameters are converted into a first feature vector, and the structured feature information in the feature information is converted into a second feature vector.
[0012] The similarity between the first feature vector and the second feature vector is used as the first matching parameter.
[0013] In one possible implementation, the step of converting the requirement parameters into a first feature vector and converting the structured feature information in the feature information into a second feature vector based on a lightweight semantic vectorization model includes:
[0014] The text content in the requirement parameters is segmented into words to obtain a first word sequence, and the text content in the structured feature information is segmented into words to obtain a second word sequence.
[0015] The first word sequence is input into the lightweight semantic vectorization model to output a first word vector set, and the first word vector set is aggregated to generate the first feature vector; and
[0016] The second word sequence is input into the lightweight semantic vectorization model to output a second set of word vectors, and the second set of word vectors is aggregated to generate the second feature vector.
[0017] In one possible implementation, the feature information further includes unstructured feature information of the corresponding candidate object;
[0018] The step of inputting the requirement parameters, the feature information, and the first matching parameters into the large language model to obtain the second matching parameters corresponding to any candidate object includes:
[0019] The required parameters, the first matching parameters, and the structured and unstructured feature information in the feature information are filled into a preset text template to obtain the model input prompt text;
[0020] The model input prompt text is input into the large language model to obtain the second matching parameter.
[0021] In one possible implementation, obtaining at least one candidate object from the information database based on the received user demand parameters includes:
[0022] According to the preset feature mapping rules, the requirement parameters are converted into at least one retrieval feature value;
[0023] The at least one search feature value is input into the search engine to obtain candidate objects that successfully match the search feature value from the information database.
[0024] In one possible implementation, the feature mapping rule includes the correspondence between constraints and retrieved feature values;
[0025] The step of converting the requirement parameters into at least one retrieval feature value according to a preset feature mapping rule includes:
[0026] The requirement parameters are semantically parsed to identify at least one constraint.
[0027] Based on the correspondence, determine the retrieval feature value corresponding to each constraint condition.
[0028] In one possible implementation, after determining the corresponding first matching parameter and before inputting the requirement parameter, the feature information, and the first matching parameter into the large language model, the method further includes:
[0029] It is determined that the first matching parameter is greater than or equal to a preset matching parameter threshold;
[0030] The method further includes:
[0031] If the first matching parameter is less than the preset matching parameter threshold, then the requirement parameter, the feature information, and the first matching parameter will not be input into the large language model.
[0032] Secondly, this application provides an apparatus for screening objects, comprising:
[0033] The receiving module is used to obtain at least one candidate object and the feature information corresponding to each candidate object from the information database based on the received user demand parameters;
[0034] The processing module is configured to, for any candidate object, determine a first matching parameter based on the requirement parameters and the feature information corresponding to the candidate object, and then input the requirement parameters, the feature information, and the first matching parameter into a large language model to obtain a second matching parameter corresponding to the candidate object. The first matching parameter represents the similarity between the feature information and the requirement parameters, and the second matching parameter represents the degree to which the candidate object matches the requirement parameters. Based on the second matching parameter corresponding to each candidate object, a target object that meets preset screening conditions is determined from the at least one candidate object.
[0035] Thirdly, this application provides an electronic device, comprising:
[0036] Memory, used to store computer programs;
[0037] A processor is configured to invoke program instructions stored in the memory and execute the steps included in the method as described in any one of the first aspects, according to the obtained program instructions.
[0038] Fourthly, this application provides a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a computer, cause the computer to perform the method as described in any one of the first aspects.
[0039] The beneficial effects of this application are as follows:
[0040] This application provides a method, apparatus, and electronic device for screening objects. First, based on received user demand parameters, at least one candidate object and its corresponding feature information are obtained from an information database. Then, for any candidate object, a first matching parameter is determined based on the demand parameters and the corresponding feature information. The demand parameters, feature information, and the first matching parameter are then input into a large language model to obtain a second matching parameter for each candidate object. Finally, based on the second matching parameter for each candidate object, target objects meeting preset screening conditions are determined from at least one candidate object. This application achieves semantic-based vectorized matching by vectorizing the demand parameters and candidate object feature information and calculating their similarity. The vectorized matching result, along with the demand parameters and feature information, is then input into a large language model to achieve deep matching decision-making combining semantic understanding and multi-dimensional analysis. This transforms the traditional screening process, which relies on subjective human judgment, into a data- and model-driven automated process, effectively improving the consistency, stability, and execution efficiency of the screening process, and significantly enhancing the objectivity and accuracy of the matching results. Attached Figure Description
[0041] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0042] Figure 1 A flowchart illustrating a method for filtering objects provided in an embodiment of this application;
[0043] Figure 2 This is a schematic diagram of another method for filtering objects provided in an embodiment of this application;
[0044] Figure 3This is a flowchart illustrating a method for filtering objects according to an embodiment of this application;
[0045] Figure 4 A schematic diagram of feature information provided in an embodiment of this application;
[0046] Figure 5 This is a schematic diagram of the structure of a device for screening objects provided in an embodiment of this application;
[0047] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0048] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0049] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0050] In business practice, it is often necessary to efficiently and accurately filter out objects that meet specific criteria from a large candidate set to support subsequent decision-making or operations. For example, in scenarios such as corporate mergers and acquisitions, talent recruitment, supplier evaluation, or initial selection of investment targets, it is necessary to identify the target that best matches the business needs from a massive number of candidates.
[0051] Currently, the industry generally adopts a screening model dominated by human experience. The typical process is as follows: First, based on the user's requirements, conditions are matched against a database using pre-defined rules to generate a preliminary candidate set. Then, this candidate set is submitted to domain experts, who manually review, subjectively adjust, and manually sort the candidates based on their personal experience, ultimately determining the target. This method relies heavily on human intervention and subjective judgment throughout the entire decision-making process.
[0052] However, this existing technology has significant drawbacks. The data dimensions relied upon in the screening process are usually relatively singular, mainly limited to structured information, making it difficult to effectively integrate and utilize unstructured data such as text. Furthermore, from initial matching to final decision, the entire process relies on human experience, resulting in screening results that are heavily influenced by personal subjectivity, leading to poor consistency and reproducibility. In addition, manual processing is inefficient, unable to handle large-scale, high-concurrency screening demands, and cannot achieve real-time response. Therefore, existing methods struggle to achieve efficient and accurate object screening while ensuring objectivity and consistency.
[0053] To address the aforementioned problems, embodiments of this application provide a method, apparatus, and electronic device for filtering objects. For ease of understanding, the method, apparatus, and electronic device for filtering objects provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.
[0054] like Figure 1 The diagram shown is a flowchart illustrating a method for filtering objects according to an embodiment of this application. The details are as follows:
[0055] S101: Based on the received user demand parameters, obtain at least one candidate object and the feature information corresponding to each candidate object from the information database;
[0056] S102: For any candidate object, after determining the first matching parameter based on the requirement parameter and the feature information corresponding to the candidate object, the requirement parameter, the feature information and the first matching parameter are input into the large language model to obtain the second matching parameter corresponding to the candidate object, wherein the first matching parameter represents the similarity between the feature information and the requirement parameter, and the second matching parameter represents the degree to which the candidate object meets the requirement parameter;
[0057] S103: Based on the second matching parameter corresponding to each candidate object, determine the target object that meets the preset screening conditions from the at least one candidate object.
[0058] The method for screening objects provided in this application firstly queries candidate objects from an information database based on demand parameters and obtains their corresponding feature information. This overcomes the limitation of existing technologies that rely solely on limited structured data, enabling the simultaneous acquisition of both structured and unstructured feature information, thus significantly expanding the data foundation for evaluation. Secondly, by semantically vectorizing and calculating similarity between the demand parameters and the feature information of each candidate object, a first matching parameter is generated. This first matching parameter reflects the initial semantic association strength between the candidate object and the demand, achieving objective and automated preliminary screening based on data similarity, replacing traditional methods based on manual rules or simple conditions. Furthermore, the demand parameters, feature information, and the first matching parameter are input into a large language model. Based on its deep semantic understanding and multi-dimensional analysis capabilities, the large language model comprehensively evaluates the input information and outputs a more accurate second matching parameter. This represents a leap from matching based on surface similarity to matching based on deep semantics and comprehensive contextual understanding, significantly reducing reliance on the subjective experience of human experts and improving the objectivity and consistency of the evaluation process. Finally, based on the second matching parameters of each candidate object and combined with preset screening conditions, the target object is determined. The entire process achieves full-link automation from information acquisition and initial screening to in-depth evaluation. While significantly improving processing efficiency, it effectively overcomes the fundamental defects of existing technologies, which rely on human judgment, resulting in strong subjectivity, low consistency, and difficulty in achieving efficient and accurate matching.
[0059] In this embodiment of the application, at least one candidate object is obtained from the information database based on the received user demand parameters. The specific process is as follows:
[0060] First, structured feature mapping rules are pre-configured, clarifying the conversion logic between natural language or semi-structured user requirements and searchable structured fields in the information database. For example, the rules can stipulate that "industry is [a certain industry]" in the requirement parameters should be mapped to the "industry chain classification" field in the information database for exact matching; "operating revenue is greater than [value A]" should be mapped to the "operating revenue" field and converted into a range query condition of "greater than or equal to value A".
[0061] Then, according to preset feature mapping rules, the demand parameters are converted into at least one search feature value. Specifically, the user-input demand parameters are semantically parsed to identify at least one constraint condition. Each constraint condition typically consists of two parts: an attribute (e.g., "industry") and a value (e.g., "semiconductor industry chain"). According to the preset feature mapping rules, each identified constraint condition is converted into a corresponding search feature value. For example, the identified constraint condition {attribute: "industry", value: "semiconductor industry chain"} is converted into the search feature value {search field: "industry_chain", query logic: "equals 'semiconductor industry chain'"} according to the rules.
[0062] Finally, the at least one search feature value is input into the search engine to retrieve candidate objects that successfully match all search feature values from the indexed information database. A successful match means that the value of the candidate object in the corresponding search field meets the conditions specified by the corresponding query logic. Simultaneously, the feature information corresponding to each candidate object is obtained, including its structured and unstructured feature information.
[0063] Through the above process, the efficient and accurate conversion from user needs to machine-executable retrieval instructions was achieved, and the preliminary recall of candidate objects and the acquisition of multi-dimensional features were completed, providing a data foundation for subsequent matching analysis.
[0064] In one embodiment, a first matching parameter for each candidate object is determined through the following steps, wherein the first matching parameter is used to characterize the preliminary semantic similarity between the feature information of the candidate object and the user's requirement parameters:
[0065] Based on a lightweight semantic vectorization model, the requirement parameters are converted into a first feature vector, and the structured feature information in the feature information is converted into a second feature vector.
[0066] First, the text content in the requirement parameters is segmented into words to obtain the first word sequence, and the text content in the structured feature information is segmented into words to obtain the second word sequence.
[0067] Specifically, the text content in the user-input parameters is segmented and converted into a first word sequence. Simultaneously, the text content in the structured feature information (e.g., fields like "Industry Classification: Manufacturing" and "Operating Revenue: 10 million") of the candidate object feature information is also segmented in the same way to generate a second word sequence.
[0068] Then, the first word order is input into a lightweight semantic vectorization model (such as the Word2Vec model), which maps each word in the word sequence to a word vector and uses all the output word vectors as the first word vector set. Subsequently, an aggregation operation is performed on the first word vector set, such as taking the arithmetic mean of each word vector in the first word vector set, to generate a first feature vector representing the entire requirement parameter.
[0069] Similarly, the second word sequence is input into the same lightweight semantic vectorization model to obtain the corresponding set of second word vectors. Then, the same aggregation operation is performed on the set of second word vectors to generate a second feature vector representing the structured features of the candidate object.
[0070] Finally, the similarity between the first feature vector and the second feature vector is calculated; for example, the cosine similarity between the first and second feature vectors is calculated. This similarity value is used as the first matching parameter for the candidate object.
[0071] The above embodiments demonstrate a preliminary matching process based on semantic vectorization. This method utilizes a lightweight semantic vectorization model to map textual information to corresponding word vectors. Through vector aggregation and similarity calculation, it can efficiently and objectively quantify the semantic relevance between candidate objects and user needs. Compared to traditional methods relying on manual rules or keyword matching, this significantly improves the automation level and processing efficiency of the preliminary screening, and ensures that the calculation of the first matching parameter is entirely algorithm-based, eliminating subjective bias caused by human intervention.
[0072] In one embodiment, the second matching parameter is determined through the following steps:
[0073] The required parameters, the first matching parameters, and the structured and unstructured feature information from the feature information are filled into a preset text template to obtain the model input prompt text;
[0074] The model input prompt text is fed into the large language model to obtain the second matching parameter.
[0075] Specifically, the requirement parameters, the first matching parameters, and the structured and unstructured feature information contained in the feature information are filled and formatted according to a preset, structured text template. The text template usually includes a clear instruction role, task description, classification of input information, and output format requirements, such as: "You are a corporate M&A consultant. Please assess the degree of matching of this company as an M&A target based on the following information..." Then, each information item is filled into the corresponding position to generate model input prompt text that conforms to the understanding habits of large language models.
[0076] Then, the input prompt text is fed into a pre-trained large language model (such as Qwen, DeepSeek, etc.). Based on its deep semantic understanding and contextual reasoning ability of the input text, the large language model combines the preliminary matching situation represented by the first matching parameter to comprehensively evaluate the structured data and unstructured information of the candidate object and outputs the corresponding matching degree evaluation result. The evaluation result can be a directly given quantitative score or a natural language description including the score conclusion. The corresponding score value is extracted from the description as the second matching parameter of the candidate object.
[0077] It should be noted that the large language model in this application includes, but is not limited to, models such as Qwen and DeepSeek, and this application does not make any specific limitations on them.
[0078] Through the above steps, deep and multi-dimensional matching analysis based on a large language model is achieved, so that the final matching results not only reflect the similarity of surface data, but also integrate semantic understanding and business logic judgment, thereby significantly improving the accuracy of target object selection and the reliability of decision-making.
[0079] In this application, after generating the second matching parameter corresponding to each candidate object, target objects that meet the preset filtering conditions are determined from the candidate objects. The filtering conditions can be configured according to actual business needs.
[0080] For example, the filtering criteria are set to compare the second matching parameter corresponding to all candidate objects with a preset filtering threshold, and select candidate objects whose second matching parameter is greater than or equal to the preset filtering threshold as target filtering objects.
[0081] Optionally, the filtering criteria can be set to sort all candidate objects according to the second matching parameters in ascending or descending order, and then select the candidate objects with the highest second matching parameters.
[0082] Through the above steps, this application realizes automated and quantitative screening and sorting from the initial candidate set to the final target object, and fully constructs a full-link intelligent decision-making process from data acquisition, preliminary matching, in-depth evaluation to result output, which significantly improves the objectivity, efficiency and usability of object screening.
[0083] In another embodiment, such as Figure 2 The diagram shown is a schematic flowchart of another method for filtering objects provided in an embodiment of this application, as detailed below:
[0084] S201: Based on the received user demand parameters, obtain at least one candidate object and the feature information corresponding to each candidate object from the information database;
[0085] S202: For any candidate object, the first matching parameter is determined based on the requirement parameters and the feature information corresponding to any candidate object;
[0086] S203: Determine whether the first matching parameter is greater than or equal to the preset matching parameter threshold; if yes, proceed to S204; if no, proceed to step S206.
[0087] S204: Input the requirement parameters, feature information and the first matching parameters into the large language model to obtain the second matching parameters corresponding to any candidate object;
[0088] S205: Based on the second matching parameter corresponding to each candidate object, determine the target object that meets the preset screening conditions from the at least one candidate object.
[0089] S206: Do not input the requirement parameters, feature information, and first matching parameters into the large language model.
[0090] In the above embodiments, by quickly filtering the preliminary matching results, the number of candidate objects that need to be called for deep processing of large language models is effectively reduced, the computational load and response latency of the model are reduced, and it is especially suitable for scenarios that process massive candidate objects, realizing intelligent allocation and saving of computing resources.
[0091] The following will use a corporate merger and acquisition scenario as an example to illustrate the above methods;
[0092] Suppose an investment institution plans to find M&A targets in the semiconductor industry chain, and its requirements are: "Industry: semiconductor industry chain; sector: manufacturing; revenue: greater than 9 million; net profit: greater than 500,000".
[0093] like Figure 3 As shown, this is a flowchart illustrating a method for filtering objects provided in an embodiment of this application. First, as... Figure 3 (A) Input the demand parameters into the search engine. Based on these parameters, construct a query statement using feature mapping rules to retrieve a list of companies from the company information database, including candidate companies and their corresponding feature information. Figure 4 The diagram shown is a schematic diagram of feature information provided in an embodiment of this application. The candidate objects include, for example, Company A and Company B. Table 1 shows the structured feature information corresponding to Company A and Company B, including industry classification, sector classification, operating income and net income. Table 2 shows the unstructured feature information corresponding to Company A and Company B, including patents, public opinion news and investment and financing information.
[0094] Then as Figure 3(B) Input the demand parameters and structured feature information from the company list into a lightweight semantic vectorization model for semantic vectorization processing, converting the text information into feature vector representations. Specifically, the demand parameters are transformed into a demand vector (i.e., the first feature vector) of [1, 2, 3, 1, 4, 5], the structured feature information of company A is transformed into a company vector (i.e., the second feature vector) of [1, 2, 3, 2, 1, 6], and the structured feature information of company B is transformed into a company vector (i.e., the second feature vector) of [2, 3, 5, 1, 2, 5]. Then, the similarity between the demand vector and each company vector is calculated, serving as the first matching parameter for each company. Taking company A as an example, the similarity is calculated using the cosine similarity formula:
[0095]
[0096] Where ai represents the i-th vector value in the demand vector, bi represents the i-th vector value in the enterprise vector, and b represents the number of values included in the demand vector or enterprise vector.
[0097] Substituting the vector of company A into the above formula, the similarity is calculated to be 0.901. Mapping the similarity to a percentage range, the first matching parameter of company A is approximately 90. Similarly, the first matching parameter of company B is calculated to be 92.
[0098] Subsequently, as Figure 3 (C) The requirement parameters, the first matching parameters for each enterprise, and the enterprise list (i.e., structured and unstructured feature information) are organized into the model input prompt text according to a preset text template and input into the large language model for deep comprehensive analysis. Based on semantic understanding and contextual reasoning, the large language model outputs the second matching parameters for each enterprise. Assume that the second matching parameter for enterprise A is 91 and the second matching parameter for enterprise B is 89.
[0099] Finally, based on the second matching parameter and preset screening conditions for each company (e.g., only companies with a second matching parameter greater than or equal to 80 are output, and they are sorted by net profit from largest to smallest), the final target objects are determined. Company A has a net profit of 1 million, and Company B has a net profit of 2 million. After sorting them in ascending order of net profit, the output target object order is Company B (89 points) and Company A (91 points).
[0100] The above embodiments demonstrate the complete process of this application in a real business scenario, from demand input, data retrieval, semantic matching, intelligent judgment to result output, reflecting the technical effects of the method in improving screening efficiency, ensuring matching accuracy, and enhancing decision-making objectivity.
[0101] Based on the same inventive concept, this application also provides a device for filtering objects. The principle of this text generation and filtering object device in solving the technical problem is similar to the above-mentioned filtering object method. The implementation of the filtering object device can refer to the implementation of the filtering object method, and the repeated parts will not be described again.
[0102] like Figure 5 The diagram shown is a structural schematic of a device for screening objects according to an embodiment of this application, comprising:
[0103] The receiving module 501 is used to obtain at least one candidate object and the feature information corresponding to each candidate object from the information database based on the received user demand parameters;
[0104] Processing module 502 is configured to, for any candidate object, determine a first matching parameter based on the requirement parameter and the feature information corresponding to the candidate object, and then input the requirement parameter, the feature information, and the first matching parameter into a large language model to obtain a second matching parameter corresponding to the candidate object. The first matching parameter represents the similarity between the feature information and the requirement parameter, and the second matching parameter represents the degree to which the candidate object matches the requirement parameter. Based on the second matching parameter corresponding to each candidate object, target objects that meet preset screening conditions are determined from the at least one candidate object.
[0105] In one embodiment, the feature information includes structured feature information of the candidate object;
[0106] The processing module 502 is specifically used for:
[0107] Based on a lightweight semantic vectorization model, the requirement parameters are converted into a first feature vector, and the structured feature information in the feature information is converted into a second feature vector.
[0108] The similarity between the first feature vector and the second feature vector is used as the first matching parameter.
[0109] In one embodiment, the processing module 502 is specifically used for:
[0110] The text content in the requirement parameters is segmented into words to obtain a first word sequence, and the text content in the structured feature information is segmented into words to obtain a second word sequence.
[0111] The first word sequence is input into the lightweight semantic vectorization model to output a first word vector set, and the first word vector set is aggregated to generate the first feature vector; and
[0112] The second word sequence is input into the lightweight semantic vectorization model to output a second set of word vectors, and the second set of word vectors is aggregated to generate the second feature vector.
[0113] In one embodiment,
[0114] The feature information also includes the unstructured feature information of the corresponding candidate object;
[0115] The processing module 502 is specifically used for:
[0116] The required parameters, the first matching parameters, and the structured and unstructured feature information in the feature information are filled into a preset text template to obtain the model input prompt text;
[0117] The model input prompt text is input into the large language model to obtain the second matching parameter.
[0118] In one embodiment, the receiving module 501 is specifically used for:
[0119] According to the preset feature mapping rules, the requirement parameters are converted into at least one retrieval feature value;
[0120] The at least one search feature value is input into the search engine to obtain candidate objects that successfully match the search feature value from the information database.
[0121] In one embodiment, the feature mapping rule includes a correspondence between constraints and retrieved feature values, and the receiving module 501 is specifically used for:
[0122] The requirement parameters are semantically parsed to identify at least one constraint.
[0123] Based on the correspondence, determine the retrieval feature value corresponding to each constraint condition.
[0124] In one embodiment, after determining the corresponding first matching parameter and before inputting the demand parameter, the feature information, and the first matching parameter into the large language model, the processing module 502 is further configured to:
[0125] It is determined that the first matching parameter is greater than or equal to a preset matching parameter threshold;
[0126] The processing module 502 is further configured to:
[0127] If the first matching parameter is less than the preset matching parameter threshold, then the requirement parameter, the feature information, and the first matching parameter will not be input into the large language model.
[0128] Based on the same inventive concept, this application also provides an electronic product. The implementation principle of the electronic product is similar to that of the method for screening objects. The specific implementation method of the electronic product can be found in the aforementioned method embodiment for screening objects. Repeated details will not be repeated.
[0129] like Figure 6 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application, comprising:
[0130] Memory 601 is used to store program instructions;
[0131] Processor 602 is configured to call program instructions stored in the memory and execute the following steps according to the obtained program instructions:
[0132] Based on the received user demand parameters, at least one candidate object and the feature information corresponding to each candidate object are obtained from the information database;
[0133] For any candidate object, after determining the first matching parameter based on the requirement parameter and the feature information corresponding to the candidate object, the requirement parameter, the feature information and the first matching parameter are input into the large language model to obtain the second matching parameter corresponding to the candidate object. The first matching parameter represents the similarity between the feature information and the requirement parameter, and the second matching parameter represents the degree to which the candidate object meets the requirement parameter.
[0134] Based on the second matching parameter corresponding to each candidate object, a target object that meets the preset filtering conditions is determined from the at least one candidate object.
[0135] Based on the same inventive concept, this application also provides a computer-readable storage medium. The implementation principle of the computer-readable storage medium is similar to that of the method for filtering objects. The specific implementation of the computer-readable storage medium can be found in the aforementioned embodiments of the method for filtering objects, and the repeated parts will not be described again.
[0136] This application provides a method, apparatus, and electronic device for screening objects. First, based on requirement parameters, it queries a database for candidate objects and obtains their corresponding feature information. This overcomes the limitation of existing technologies that rely solely on limited structured data, enabling the simultaneous acquisition of both structured and unstructured feature information, thus significantly expanding the data foundation for evaluation. Second, by semantically vectorizing and calculating similarity between the requirement parameters and the feature information of each candidate object, a first matching parameter is generated. This first matching parameter reflects the initial semantic association strength between the candidate object and the requirement, achieving objective and automated preliminary screening based on data similarity, replacing traditional methods based on manual rules or simple conditions. Further, the requirement parameters, feature information, and the first matching parameter are input into a large language model. Based on its deep semantic understanding and multi-dimensional analysis capabilities, the large language model comprehensively evaluates the input information and outputs a more accurate second matching parameter. This represents a leap from matching based on surface similarity to matching based on deep semantics and comprehensive contextual understanding, significantly reducing reliance on the subjective experience of human experts and improving the objectivity and consistency of the evaluation process. Finally, based on the second matching parameters of each candidate object and combined with preset screening conditions, the target object is determined. The entire process achieves full-link automation from information acquisition and initial screening to in-depth evaluation. While significantly improving processing efficiency, it effectively overcomes the fundamental defects of existing technologies, which rely on human judgment, resulting in strong subjectivity, low consistency, and difficulty in achieving efficient and accurate matching.
[0137] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0138] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0139] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0140] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0141] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. A method for filtering objects, characterized in that, include: Based on the received user demand parameters, at least one candidate object and the feature information corresponding to each candidate object are obtained from the information database; For any candidate object, after determining the first matching parameter based on the requirement parameter and the feature information corresponding to the candidate object, the requirement parameter, the feature information and the first matching parameter are input into the large language model to obtain the second matching parameter corresponding to the candidate object. The first matching parameter represents the similarity between the feature information and the requirement parameter, and the second matching parameter represents the degree to which the candidate object meets the requirement parameter. Based on the second matching parameter corresponding to each candidate object, a target object that meets the preset filtering conditions is determined from the at least one candidate object.
2. The method according to claim 1, characterized in that, The feature information includes the structured feature information of the candidate object; The step of determining the first matching parameter of any candidate object based on the requirement parameters and the feature information corresponding to any candidate object includes: Based on a lightweight semantic vectorization model, the requirement parameters are converted into a first feature vector, and the structured feature information in the feature information is converted into a second feature vector. The similarity between the first feature vector and the second feature vector is used as the first matching parameter.
3. The method according to claim 2, characterized in that, The lightweight semantic vectorization model, which converts the requirement parameters into a first feature vector and the structured feature information in the feature information into a second feature vector, includes: The text content in the requirement parameters is segmented into words to obtain a first word sequence, and the text content in the structured feature information is segmented into words to obtain a second word sequence. The first word sequence is input into the lightweight semantic vectorization model to output a first word vector set, and the first word vector set is aggregated to generate the first feature vector; and The second word sequence is input into the lightweight semantic vectorization model to output a second set of word vectors, and the second set of word vectors is aggregated to generate the second feature vector.
4. The method according to claim 2, characterized in that, The feature information also includes the unstructured feature information of the corresponding candidate object; The step of inputting the requirement parameters, the feature information, and the first matching parameters into the large language model to obtain the second matching parameters corresponding to any candidate object includes: The required parameters, the first matching parameters, and the structured and unstructured feature information in the feature information are filled into a preset text template to obtain the model input prompt text; The model input prompt text is input into the large language model to obtain the second matching parameter.
5. The method according to claim 1, characterized in that, The step of retrieving at least one candidate object from the information database based on the received user demand parameters includes: According to the preset feature mapping rules, the requirement parameters are converted into at least one retrieval feature value; The at least one search feature value is input into the search engine to obtain candidate objects that successfully match the search feature value from the information database.
6. The method according to claim 5, characterized in that, The feature mapping rules include the correspondence between constraints and retrieved feature values; The step of converting the requirement parameters into at least one retrieval feature value according to a preset feature mapping rule includes: The requirement parameters are semantically parsed to identify at least one constraint. Based on the correspondence, determine the retrieval feature value corresponding to each constraint condition.
7. The method according to any one of claims 1 to 6, characterized in that, After determining the corresponding first matching parameter, and before inputting the requirement parameter, the feature information, and the first matching parameter into the large language model, the method further includes: It is determined that the first matching parameter is greater than or equal to a preset matching parameter threshold; The method further includes: If the first matching parameter is less than the preset matching parameter threshold, then the requirement parameter, the feature information, and the first matching parameter will not be input into the large language model.
8. A device for screening objects, characterized in that, include: The receiving module is used to obtain at least one candidate object and the feature information corresponding to each candidate object from the information database based on the received user demand parameters; The processing module is configured to, for any candidate object, determine a first matching parameter based on the requirement parameters and the feature information corresponding to the candidate object, and then input the requirement parameters, the feature information, and the first matching parameter into a large language model to obtain a second matching parameter corresponding to the candidate object. The first matching parameter represents the similarity between the feature information and the requirement parameters, and the second matching parameter represents the degree to which the candidate object matches the requirement parameters. Based on the second matching parameter corresponding to each candidate object, a target object that meets preset screening conditions is determined from the at least one candidate object.
9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to invoke program instructions stored in the memory and execute the steps included in the method as described in any one of claims 1-7 according to the obtained program instructions.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1-7.