Object searching method and device, computer equipment and storage medium

Abstract

The embodiment of the invention relates to an object searching method and device, computer equipment and a storage medium, and the method comprises the steps: searching a target department corresponding to an object searching request from a knowledge base when the object searching request is obtained; for each object in the target department, obtaining a target and key result data from a knowledge base; and according to the target of each object and the key result data, performing a context reverse search operation so as to determine a target object from the plurality of objects through the context reverse search operation. Therefore, the related departments can be retrieved from the knowledge base, then the data of the objects in the departments are obtained, and finally the target object is determined through the context reverse search operation, a traditional search personnel mode is not needed, the searched object is determined in a mode of combining knowledge base retrieval and context reverse search, and the efficiency of the search personnel is improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the technical field of natural language processing, and in particular, to an object search method, apparatus, computer device, and storage medium. Background Art

[0002] In enterprise-level technical collaboration, there are serious efficiency problems in cross-departmental technical consultation and project team formation matching. Currently, when looking for technical support personnel within an enterprise or forming a project team by looking for personnel, it mainly relies on manual recommendation or a traditional natural language recommendation engine to query a predetermined number of recommenders with the highest degree of association with the search or recruitment information as the personnel required for the search or project team formation.

[0003] Traditional natural language engines have limited language understanding capabilities, and multiple levels of transmission are required to determine an object. Moreover, information attenuation is serious during multiple rounds of retelling, which easily leads to a low matching degree between the determined personnel and the required personnel, resulting in the problem of repeated searches, and further leading to a long search time and low efficiency for search personnel. Therefore, how to improve the efficiency of finding personnel in an enterprise has become an urgent problem to be solved. Summary of the Invention

[0004] In view of this, to solve the above technical problems or some of the technical problems, embodiments of the present invention provide an object search method, apparatus, computer device, and storage medium.

[0005] In a first aspect, embodiments of the present invention provide an object search method, including:

[0006] When an object search request is obtained, retrieve the target department corresponding to the object search request from a knowledge base, where the knowledge base includes: the goals and key results data of each department, and the goals and key results data of the objects included in each department;

[0007] For each object in the target department, obtain the goals and key results data from the knowledge base;

[0008] Perform a context reverse search operation based on the goals and key results data of each object, so as to determine a target object from multiple objects through the context reverse search operation.

[0009] In a possible implementation manner, before obtaining the object search request, the method further includes:

[0010] Obtain the goals and key results texts corresponding to multiple departments, and obtain the goals and key results texts corresponding to the objects included in each department, to obtain a goals and key results set;

[0011] Perform invalid data filtering processing, technical entity recognition processing, and semantic disambiguation processing on the goals and key results set;

[0012] Vectorize and encode the processed objectives and key results set to obtain the objective and key result data for each of the departments and the objective and key result data for each of the objects;

[0013] Generate the knowledge base from the objective and key result data for each of the departments and the objective and key result data for each of the objects.

[0014] In one possible implementation, the method further includes:

[0015] When it is detected that any of the objective and key result texts is updated, update the objective and key result data in the knowledge base according to the update content.

[0016] In one possible implementation, the method further includes:

[0017] When it is detected that any of the objective and key result data in the knowledge base has not been updated for a preset duration, adjust the time decay factor of the unupdated objective and key result data.

[0018] In one possible implementation, retrieving the target department corresponding to the object lookup request from the knowledge base includes:

[0019] After semantic parsing of the object lookup request, perform intent recognition to obtain a recognition result;

[0020] Input the recognition result into a trained large model to retrieve the target department corresponding to the recognition result from the knowledge base through the large model.

[0021] In one possible implementation, the context reverse lookup operation based on the objective and key result data for each of the objects includes:

[0022] Optimize the prompt words of the large model according to the object lookup request and the objective and key result data of all objects in the target department to obtain new prompt words;

[0023] Enable the large model to perform a context reverse lookup operation on the objects in the target department according to the new prompt words to obtain the target object.

[0024] In one possible implementation, after determining the target object from multiple objects through the context reverse lookup operation, the method further includes:

[0025] Obtain explicit feedback and / or implicit feedback for the target object, where the explicit feedback includes: the satisfaction level for the target object, and the implicit feedback includes: the referral behavior for the target object;

[0026] Optimize the large model according to the explicit feedback and / or the implicit feedback, where the explicit feedback is used to optimize the weight configuration of the large model for the target object, and the implicit feedback is used to optimize the object recommendation strategy of the large model.

[0027] In a possible implementation, after determining the target object from multiple objects through the context reverse lookup operation, the method further includes:

[0028] Send a confirmation message to the target object;

[0029] When a response message returned by the target object for the confirmation message is not received within a preset duration, determine a new target object corresponding to the object lookup request.

[0030] In a second aspect, an embodiment of the present invention provides an object lookup device, including:

[0031] A retrieval module, configured to retrieve a target department corresponding to the object lookup request from a knowledge base when an object lookup request is obtained, where the knowledge base includes: target and key result data of each department, and target and key result data of objects included in each department;

[0032] An acquisition module, configured to acquire target and key result data from the knowledge base for each object in the target department;

[0033] A determination module, configured to perform a context reverse lookup operation according to the target and key result data of each object, so as to determine a target object from multiple objects through the context reverse lookup operation.

[0034] In a third aspect, an embodiment of the present invention provides a computer device, including: a processor and a memory, where the processor is configured to execute an object lookup program stored in the memory to implement the object lookup method according to any one of the first aspects above.

[0035] In a fourth aspect, an embodiment of the present invention provides a storage medium, where the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the object lookup method according to any one of the first aspects above.

[0036] The object search solution provided by the embodiments of the present invention, when an object search request is obtained, retrieves the target department corresponding to the object search request from a knowledge base, where the knowledge base includes: the objectives and key results data of each department, and the objectives and key results data of the objects included in each department; for each object in the target department, obtains the objectives and key results data from the knowledge base; performs a context reverse search operation according to the objectives and key results data of each object, so as to determine the target object from multiple objects through the context reverse search operation. Thus, the relevant department can be retrieved from the knowledge base, then the data of each object in the department can be obtained, and finally the target object can be determined through the context reverse search operation. Without using the traditional way of searching for personnel, the target object to be searched can be accurately determined by combining knowledge base retrieval and context reverse search, thereby improving the efficiency of searching for personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic flowchart of an object search method provided by an embodiment of the present invention;

[0038] Figure 2 It is a schematic flowchart of another object search method provided by an embodiment of the present invention;

[0039] Figure 3 It is a schematic flowchart of still another object search method provided by an embodiment of the present invention;

[0040] Figure 4 It is a schematic flowchart of data processing in an object search method provided by an embodiment of the present invention;

[0041] Figure 5 It is a schematic flowchart of yet another object search method provided by an embodiment of the present invention;

[0042] Figure 6 It is a schematic structural diagram of an object search device provided by an embodiment of the present invention;

[0043] Figure 7 It is a schematic structural diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0045] For the convenience of understanding the embodiments of the present invention, the following will further explain with specific embodiments in conjunction with the accompanying drawings. The embodiments do not constitute a limitation on the embodiments of the present invention.

[0046] Figure 1 The following is a schematic flowchart of an object search method provided by an embodiment of the present invention. As Figure 1 shown, the method specifically includes:

[0047] S11. When an object search request is obtained, retrieve the target department corresponding to the object search request from the knowledge base. The knowledge base includes: the objectives and key results data of each department, and the objectives and key results data of the objects included in each department.

[0048] The object search method provided by the embodiment of the present invention is applied to a computer device, which may include but is not limited to: servers, desktop computers, tablet computers, etc. It is specifically applied to cross-departmental technical consultation and personnel search in enterprise-level technical collaboration, and the application scenario of finding technical personnel in an enterprise for project team matching. Specifically, relevant departments are retrieved from the knowledge base, then the objectives and key results data of each object within the department are obtained, and finally the target object is determined through context reverse search operations.

[0049] In this embodiment, a knowledge base at the department and personnel levels within the department is established in advance. The knowledge base includes but is not limited to: enterprise Objectives and Key Results (OKR) data, organizational structure, expert skills, historical technical consultation records, etc. Use a pre-trained large model (such as BERT, GPT, etc.) to convert the obtained text data into vectors and store them in the knowledge base. For example, the text is: K8s operation and maintenance expert, and the vector representation is: [0.12, 0.85, -0.33,...] (the number of dimensions depends on the model).

[0050] Obtain the user's technical consultation request as an object search request and convert it into a retrievable vector. For example, the user enters: I need an engineer who is good at Kubernetes cluster tuning. First, entity recognition can be performed on the user input: extract keywords (such as Kubernetes, cluster tuning, engineer). Then, semantic parsing is performed to determine the query type (such as skill matching, technical consultation, project teaming). Then, vectorization processing is performed to convert the query into a vector for similarity retrieval.

[0051] Furthermore, retrieve the target department according to the data in the knowledge base to determine which department best meets the query requirements. Calculate the similarity between the vector of the user's object query request and the department OKR vector to find the most matching target department.

[0052] The matching method can be cosine similarity calculation to measure the similarity between the query vector and the department vector. Set a minimum matching score (confidence threshold (θ = 0.85)) to filter the most relevant departments. The most relevant N departments can be returned, and the department with the highest matching degree can be selected.

[0053] S12. For each object in the target department, obtain the target and key result data from the knowledge base.

[0054] In this embodiment, after determining the target department, it is necessary to further extract the target and key result data of each object (expert, member) within the department for precise matching in the next step.

[0055] Specifically, the target department contains multiple members, each member is an object, and the target and key result data of each object are stored in the knowledge base. The unique identification ID of all members of the department can be queried from the knowledge base. Using the member ID as an index, query the knowledge base to obtain the target and key result data of the member.

[0056] After obtaining the target and key result data, it needs to be formatted and converted into a structure suitable for large model inference (for example, JSON structure).

[0057] S13. Perform a context reverse lookup operation based on the target and key result data of each object to determine the target object from multiple objects through the context reverse lookup operation.

[0058] In this embodiment, use the target and key result data of each object in the target department to perform a context reverse lookup to screen out the target object that best meets the user's needs from multiple candidate objects. The main purpose of the context reverse lookup is to improve the accuracy of the matching and ensure that the recommended personnel not only meet the requirements in terms of technical capabilities but also are closest to the user's problem in terms of actual work goals and results.

[0059] Specifically, use the target and key result data of each object to construct a prompt for the large model, and input it into the large model for secondary screening.

[0060] Prompt structure example:

[0061] User's consulting question: "How to optimize the resource scheduling of the Kubernetes cluster and reduce the node load?". Target department: Cloud computing team. The members of this department and their OKR data are as follows:

[0062] 1. Zhang San: Objective (O): Improve the stability of the Kubernetes cluster. Key Result (KR): The average failure rate of the cluster < 0.5%, and handle more than 50 container optimization tasks.

[0063] 2. Li Si: Objective (O): Improve DevOps automated deployment. Key Result (KR): DevOps deployment efficiency increased by 20%, and more than 10 CI / CD pipelines added.

[0064] Please analyze which of the above candidates best meets the requirements based on the user's question and return the best-matched candidate.

[0065] Furthermore, call the large model for context reverse lookup. Use the constructed Prompt as input and call the large model for reasoning. The large model will combine the user's question and the OKR semantic information of multiple objects within the target department, score each object, and determine the target object as the best recommended candidate. The information of the target object can be returned to the questioning user, or the questioning user's information and the question can be sent to the target object.

[0066] As an example, Figure 2 It is a schematic flowchart of another object search method provided by the embodiment of the present invention.

[0067] First, perform OKR data processing, extract data from the enterprise OKR data source. Conduct structured processing through the data cleaning module to filter invalid information. Build a knowledge base and store it in the distributed vector database through the vectorization engine. Then, perform user question parsing, obtain the technical consultation question input by the user, identify technical terms and intentions through the semantic parser as the object search request, and conduct the first-round retrieval through the hybrid reasoning engine to obtain the target department. Obtain the OKR data of the objects within the target department (perform semantic comparison using the OKR data), optimize the prompt according to the department relationship graph (optimize the match based on the organizational structure) and the OKR data of each object, generate a new prompt, enhance the large model's understanding ability, and conduct context reverse lookup. Combine the optimized data and call the large model for precise matching. The secondary confirmation routing ensures that the recommended result meets the business requirements (such as timeout re-routing). Collect user feedback through the feedback learning module, optimize the recommendation strategy, enable the system to self-improve, can achieve efficient person search, reduce information transmission loss, improve accuracy, and the feedback learning mechanism continuously optimizes the matching effect and improves the collaborative efficiency of technical resources within the enterprise.

[0068] The object search method provided by the embodiments of the present invention includes, when an object search request is obtained, retrieving a target department corresponding to the object search request from a knowledge base, where the knowledge base includes: the objectives and key results data of each department, and the objectives and key results data of the objects included in each department; for each object in the target department, obtaining the objectives and key results data from the knowledge base; performing a context reverse search operation according to the objectives and key results data of each object, so as to determine a target object from multiple objects through the context reverse search operation. Thus, relevant departments can be retrieved from the knowledge base, then the data of each object within the department can be obtained, and finally the target object can be determined through the context reverse search operation. There is no need to use the traditional method of searching for personnel. The target object to be searched can be accurately determined through knowledge base retrieval and context reverse search, thereby improving the efficiency of searching for personnel.

[0069] Figure 3 is a schematic flowchart of another object search method provided by the embodiments of the present invention, as Figure 3 shown, the method specifically includes:

[0070] S21. Obtain the objectives and key results texts corresponding to multiple departments, and obtain the objectives and key results texts corresponding to the objects included in each department, to obtain an objectives and key results set; perform invalid data filtering processing, technical entity recognition processing, and semantic disambiguation processing on the objectives and key results set;

[0071] S22. Vectorize and encode the processed objectives and key results set to obtain the objectives and key results data of each department and the objectives and key results data of each object; generate a knowledge base from the objectives and key results data of each department and the objectives and key results data of each object.

[0072] In this embodiment, the objectives (O) and key results (KR) of each department are extracted from the enterprise OKR system or database to form a department-level OKR set.

[0073] For example, for the Cloud Computing department, the objective (O): Improve the stability of the Kubernetes cluster, and the key result (KR): Reduce the failure rate to <0.5%, and process more than 50 container optimization tasks.

[0074] Further extract the OKRs of all members within each department to form an individual-level OKR set.

[0075] For example, Zhang San (Cloud Computing department), the objective (O): Improve the Kubernetes resource scheduling efficiency, and the key result (KR): Increase the CPU resource utilization rate to 80%, and optimize more than 20 scheduling policies.

[0076] Take the department-level OKR set and the individual-level OKR set as the objective and key results set. And perform preprocessing, which may include but is not limited to: technology stack extraction (for example, extracting technology stack tags from OKR text through an OKR feature extractor), invalid data filtering, technology entity recognition processing, and semantic disambiguation processing. The invalid data filtering process can be to eliminate OKRs with missing key data, such as: the KR is empty, the KR is too general (such as "improve team efficiency"), and the elimination method can be based on regular rule matching and statistical threshold elimination.

[0077] The technology entity recognition process is used to identify technology keywords in the OKR text and establish a standardized representation. For example, "K8s" → "Kubernetes", "AI model" → "deep learning model". Specifically, keyword extraction can be based on a domain dictionary and named entity recognition (NER).

[0078] The semantic disambiguation process is used to solve the problem of polysemy and ensure the correct context. For example, "cloud native" refers to Kubernetes in the container technology field and Serverless in the cloud computing infrastructure field. Specifically, the actual semantics of the current scenario can be determined by combining a knowledge graph and semantic similarity calculation (BERT cosine similarity).

[0079] Furthermore, the processed data is stored in a vectorized manner. Using a vectorization model (for example, BERT, SBERT, GPT Embeddings), the text is converted into a numerical vector for easy similarity calculation. The vector data is stored in a distributed vector database (for example, FAISS, Milvus, Weaviate) to build an enterprise OKR knowledge base. The storage structure of the knowledge base: department-level vector library (stores department-level OKR vectors), individual-level vector library (stores OKR vectors of each object), relational database (for example, MySQL, Neo4j): stores structured data, such as department-personnel relationships, historical consultation records, etc.

[0080] As an example, as Figure 4 shown in the schematic flowchart of data processing in an object search method provided by an embodiment of the present invention. After obtaining the department of the enterprise and the OKR information of the objects in each department, use regular rules to eliminate OKRs with missing skills, perform technology entity recognition based on domain dictionary matching of technology keywords, and perform semantic disambiguation processing. Convert the OKR information into vectorized storage and keep the timeliness of the knowledge base through a dynamic update algorithm.

[0081] In a possible implementation, when any objective and key result text is detected to be updated, the objective and key result data in the knowledge base is updated according to the updated content; when it is detected that any objective and key result data in the knowledge base has not been updated for more than a preset duration, the time decay factor of the unupdated objective and key result data is adjusted.

[0082] Specifically, when the objectives and key results in the department OKR or personal OKR are updated in text, the update mechanism is triggered to synchronize the latest text content of the relevant objectives and key results to the knowledge base. If any objective and key result data in the knowledge base exceeds the preset non-update time (e.g., 30 days, 60 days, one year, etc.), the system should mark it as "expired" or "invalid". To enable the system to adjust the objectives and key results according to the degree of data invalidation, the system can set a decay factor (e.g., apply a 0.7 decay factor to OKR data that has not been updated for more than one year). The unupdated data will gradually reduce its weight and relevance, thereby reducing its priority in the matching and retrieval process.

[0083] Adjustment algorithm: The decay strategy can be designed according to business requirements. For example: Simple time decay: Every once in a while, the influence of the data decreases by a fixed percentage. Dynamic decay: According to the work dynamics of the department and individuals, the decay factor of the unupdated data will be adjusted according to the urgency or importance of the task. This can greatly improve the efficiency of cross-departmental collaboration and personnel matching, while ensuring that the data always remains up-to-date and relevant, and avoiding information obsolescence affecting decision-making and execution.

[0084] S23. When an object search request is obtained, after semantic parsing of the object search request, intention recognition is performed to obtain an identification result; the identification result is input into a trained large model to retrieve the target department corresponding to the identification result from the knowledge base.

[0085] In this embodiment, the search request submitted by the user can be related information about a certain department or individual (e.g., "looking for personnel responsible for product R & D" or "team in need of technical support"). First, the semantic parsing of the user's search request needs to be performed. This process includes technical means such as word segmentation, named entity recognition (NER), and dependency syntactic analysis to extract key information in the request, etc. Key entity extraction: For example, "product R & D" and "technical support" can be used as entities to represent the department or function of the search object. Context analysis: Analyze the context of the information provided by the user to determine whether additional supplementation or context processing is required. Further, a deep learning model (such as BERT, GPT, etc.) can be used for intention classification, analyzing information such as action verbs and request targets in the text, and mapping the user's query to a specific intention label (e.g., "department search", "position search", or "role search", etc.).

[0086] Input the result of intention recognition into a trained large model. The large model queries the knowledge base based on the recognized intention and the query content to retrieve the matching target department.

[0087] S24. For each object in the target department, obtain the objective and key result data from the knowledge base.

[0088] In this embodiment, similar to step S12, specific reference can be made to Figure 1 the relevant content. For the sake of concise description, it will not be elaborated here.

[0089] S25. Optimize the prompt words of the large model according to the object search request and the objective and key result data of all objects in the target department to obtain new prompt words; so that the large model performs a context reverse search operation on the objects in the target department according to the new prompt words to obtain the target object.

[0090] In this embodiment, by comparing the search request with the objective and key result data, it is determined how to optimize the prompt words of the large model. For example, "person in charge" in the search request may correspond to the role of "department head" in the target department, and "product design" may be related to the "design team". Construct new prompt words: According to the user's query intention and the context of the target department, construct new prompt words to ensure that the large model can understand and generate relevant search contexts. Example of the original prompt word: If the query is "looking for the person in charge of product design", example of the optimized prompt word: By combining the objective and key result data of the target department, the prompt word is optimized to: "The leader in charge of product design, with experience in managing the design team, and the objectives and key results include improving product design quality and teamwork, etc.".

[0091] Input the optimized prompt words into the trained large model. The large model can understand the context and generate relevant search suggestions or results. The large model uses the optimized prompt words to perform a reverse search operation in the knowledge base to find context information related to the objects in the target department. The reverse search operation here is not just a simple keyword matching, but a deep understanding based on the context. For example, the model can extract objects related to the query (such as relevant person in charge, key positions, etc.) from the objective and key result data of the target department. The large model can also perform multi-dimensional matching based on information such as tasks, responsibilities, and key results, so as to return the most relevant target objects (such as department head, design person in charge, etc.). If necessary, further information about the target object can also be returned, such as specific responsibilities, performance objectives, contact persons, etc.

[0092] S26. Obtain explicit feedback and / or implicit feedback for the target object; optimize the large model according to the explicit feedback and / or implicit feedback. The explicit feedback is used to optimize the weight configuration of the large model for the target object, and the implicit feedback is used to optimize the object recommendation strategy of the large model.

[0093] In this embodiment, the explicit feedback is the evaluation information directly provided by the user, mainly manifested as: the satisfaction score of the target object: the user can express the satisfaction with the target object (such as a certain department or person in charge) by quantifying the score (such as 1 to 5 stars) or by a simple "like". The user's satisfaction with the target object can be directly asked through a questionnaire or survey, which may include ratings of multiple dimensions (such as response speed, work quality, collaboration effect, etc.). The implicit feedback is the satisfaction indirectly reflected by the user's behavior, mainly through the referral behavior: when the user is satisfied with a certain target object, they may recommend it to others. This behavior can be captured by tracking the user's referral operations (such as forwarding, sharing links, recommending others to join a certain department, etc.). Frequent interaction behavior: The user frequently interacts with a certain target object (such as frequently searching for the same department, having multiple exchanges with a specific person in charge, etc.).

[0094] In this embodiment, the large model is optimized through a dual-channel feedback mechanism. By the explicit feedback provided by the user, a weight or score value can be assigned to the target object. This score will directly affect the weight configuration of the target object in the model. For example: if the target object receives a high score, the model will enhance the relevance weight of this object and increase its priority in subsequent queries. If the target object receives a low score or negative feedback, the system will reduce the weight of this object, making it gradually weaken in the model recommendation.

[0095] When the user recommends a certain target object to others, the system should identify this referral behavior and make fine-tuning. Specifically, record which target objects are frequently referred, and evaluate the relevance between the referral behavior and the target object. If a target object receives more referrals, the system will make fine-tuning according to the number of referrals or the frequency of the behavior, strengthen the recommendation through positive feedback, and reward those target objects with higher referral behavior, so that this target object can obtain more priority displays in subsequent queries.

[0096] For the objects that are rejected or not recommended, use the negative feedback mechanism to reduce their priority in future recommendations. This can be captured through "rejection behavior" or "ignoring behavior".

[0097] The feedback data (explicit and implicit) will continuously be input into the system, and the recommendation strategy will be optimized through periodic model iteration. This ensures that the model can continuously learn and adapt to the changing needs of users. The dual-channel feedback mechanism can achieve the optimization of the large model. More precisely, adjust the priority of the target object in the large model to make it more in line with the actual needs of the enterprise and the user's behavior pattern.

[0098] S27. Send a confirmation message to the target object; when a response message returned by the target object for the confirmation message is not received within a preset duration, determine a new target object corresponding to the object search request.

[0099] In this embodiment, in the object search request scenario, using the secondary confirmation protocol can ensure the correctness of the target object and the timeliness of the response. Through the timeout re-routing mechanism and the referral chain length limit policy, the matching and confirmation process of the target object can be effectively optimized.

[0100] Specifically, after the system sends a confirmation message to the target object, if a response from the target object is not received within a preset duration (e.g., 2 hours), the re-routing mechanism is triggered, and the system will re-determine a new target object. This mechanism ensures that the object search request can continue to progress even when the target object fails to respond in a timely manner, avoiding the long-term stagnation of the search request.

[0101] Set the referral chain length limit policy (e.g., maximum hop count = 3). This policy limits the level or length of the referral chain, ensuring that the selection of the target object is not affected by excessive referral operations and avoiding the uncertainty brought by over-reliance on multiple referrals. If the target object fails to respond, the system will pass the request to the referrer or relevant personnel of the target object through the referral chain until the maximum hop count is reached. Each referral requires the target object to confirm and respond within a specified time. When the referral hop count reaches the maximum value, the system will stop searching through the referral chain and instead use other methods or the re-routing mechanism to find a new target object. This can avoid the problems caused by excessive delay and complex chains, enabling the object search request to efficiently and accurately find a suitable target object.

[0102] As an example, as Figure 5 shown is a flowchart of another object search method provided by an embodiment of the present invention, which can be used for model training and workflow design in a large model platform (The Hanhai LargeModel Platform (HHLM Platform) is a one-stop large model training and inference service platform for enterprise developers. It supports core capabilities such as data management, one-click fine-tuning, inference optimization, and online evaluation, helping to implement the generative AI application requirements in various industries)

[0103] The visualization and orchestration function of the large model platform allows users to easily construct and configure each processing stage in the workflow by dragging and dropping components, and each stage can be dynamically configured by dragging and dropping components. Each component corresponds to a task or operation. At the same time, dynamic routing is set up to automatically switch the processing path based on the confidence threshold (θ = 0.85). The system can judge whether to switch to other paths according to the output confidence of each stage. At the same time, a hot plug mechanism is set up to support replacing the large model instance without interrupting the service, without the need to shut down or interrupt the service. This means that different models or versions can be switched according to needs during the operation of the workflow to adapt to changing requirements or optimized models.

[0104] First, obtain the user's original question input, perform semantic parsing through a semantic parser. After success, perform intent recognition. The recognition result is input into the hybrid inference engine to retrieve the knowledge base. First, retrieve the knowledge base through the large model to obtain the target department most relevant to the user's question, extract the OKR data under the department, optimize the prompt of the large model, and use the optimized prompt to perform a reverse search once to find the most suitable target object. If the target object does not respond, secondary re-routing can be performed. At the same time, the large model can be optimized according to user feedback, and the knowledge base can be updated in real time. Thus, on the large model platform, through visualization and orchestration, dynamic routing, and the hot plug mechanism, a flexible and efficient workflow can be constructed. The dynamic routing mechanism automatically switches paths based on confidence to ensure the accuracy and real-time nature of the workflow; the hot plug mechanism ensures that the service is not interrupted during model updates, improving the availability and elasticity of the workflow.

[0105] The object search method provided by the embodiment of the present invention pre-constructs an OKR knowledge base of departments and personnel and updates it in real time. When a query object request is received, relevant departments are retrieved from the knowledge base, then the OKR data of each object within the department is obtained, and finally the target object is determined through a context reverse search operation. There is no need to use the traditional method of finding personnel. Combining the composite inference mechanism of knowledge base retrieval and context reverse search, the balance between the prompt length and the matching accuracy is achieved through two-stage processing. This architecture solves the limitations of traditional methods in semantic understanding and improves the matching accuracy. Using OKR data for vectorized storage and real-time update solves the problem of lagging knowledge update in traditional methods. A complete feedback closed-loop is formed through a dual-channel feedback mechanism to continuously optimize the recommendation strategy. The data quality of the knowledge base is improved through steps such as invalid data filtering, technical entity recognition, semantic disambiguation, and time decay. The matching accuracy is improved through the secondary confirmation routing protocol. In summary, this embodiment significantly improves the efficiency and accuracy of finding people for technical consulting, as well as the matching efficiency of project team formation.

[0106] Figure 6 For the structural schematic diagram of an object search device provided by the embodiment of the present invention, asFigure 6 As shown, the device specifically includes:

[0107] A retrieval module 31, configured to retrieve a target department corresponding to the object search request from a knowledge base when an object search request is obtained. The knowledge base includes: objective and key result data for each department, and objective and key result data for the objects included in each department;

[0108] An acquisition module 32, configured to acquire objective and key result data from the knowledge base for each object in the target department;

[0109] A determination module 33, configured to perform a context inverse search operation based on the objective and key result data of each object, so as to determine a target object from multiple objects through the context inverse search operation.

[0110] In a possible implementation manner, the acquisition module is further configured to acquire objective and key result texts corresponding to multiple departments, and acquire objective and key result texts corresponding to the objects included in each department, to obtain an objective and key result set;

[0111] A processing module 34, configured to perform invalid data filtering processing, technical entity recognition processing, and semantic disambiguation processing on the objective and key result set;

[0112] Perform vector quantization encoding on the processed objective and key result set to obtain objective and key result data for each department and objective and key result data for each object;

[0113] Generate the knowledge base with the objective and key result data for each department and the objective and key result data for each object.

[0114] In a possible implementation manner, the update module 35 is configured to update the objective and key result data in the knowledge base according to the update content when it is detected that any of the objective and key result texts is updated.

[0115] In a possible implementation manner, the update module is further configured to adjust the time decay factor of the unupdated objective and key result data when it is detected that any objective and key result data in the knowledge base has not been updated for more than a preset duration.

[0116] In a possible implementation manner, the retrieval module is specifically configured to perform semantic parsing on the object search request and then perform intent recognition to obtain a recognition result;

[0117] Input the recognition result into a trained large model, so as to retrieve the target department corresponding to the recognition result from the knowledge base through the large model.

[0118] In a possible implementation, the determining module is specifically configured to optimize the prompt words of the large model according to the object search request and the objective and key result data of all objects in the target department, so as to obtain new prompt words;

[0119] so that the large model performs a context reverse search operation on the objects in the target department according to the new prompt words to obtain the target object.

[0120] In a possible implementation, the obtaining module is further configured to obtain explicit feedback and / or implicit feedback for the target object, where the explicit feedback includes: the satisfaction degree for the target object, and the implicit feedback includes: the referral behavior for the target object;

[0121] The optimization module 36 is configured to optimize the large model according to the explicit feedback and / or the implicit feedback. The explicit feedback is used to optimize the weight configuration of the large model for the target object, and the implicit feedback is used to optimize the object recommendation strategy of the large model.

[0122] In a possible implementation, the determining module is further configured to send a confirmation message to the target object;

[0123] When a response message returned by the target object for the confirmation message is not received within a preset duration, determine a new target object corresponding to the object search request.

[0124] The device provided in this embodiment may be a device as shown in Figure 6 and can execute all steps of the method as shown in Figures 1-3 to achieve the technical effects of the method as shown in Figures 1-3 For specific reference, please refer to Figures 1-3 the relevant description. For the sake of concise description, it will not be elaborated here.

[0125] Figure 7 FIG. is a schematic structural diagram of a computer device provided in an embodiment of the present invention. Figure 7 The computer device 400 shown includes: at least one processor 401, a memory 402, at least one network interface 404, and other user interfaces 403. Each component in the electronic device 400 is coupled together through a bus system 405. It can be understood that the bus system 405 is used to realize the connection and communication between these components. The bus system 405 includes not only a data bus, but also a power bus, a control bus, and a status signal bus. However, for the sake of clear description, in Figure 7 all buses are labeled as the bus system 405.

[0126] Among them, the user interface 403 may include a display, a keyboard, or a pointing device (such as a mouse, a trackball, a touchpad, or a touch screen, etc.).

[0127] It can be understood that the memory 402 in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synch link dynamic random access memory (SLDRAM), and direct rambus random access memory (DRRAM). The memory 402 described herein is intended to include but not be limited to these and any other suitable types of memory.

[0128] In some embodiments, the memory 402 stores the following elements, executable units, or data structures, or subsets thereof, or extended sets thereof: an operating system 4021 and application programs 4022.

[0129] Among them, the operating system 4021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application programs 4022 include various application programs, such as a media player and a browser, etc., for implementing various application services. The program for implementing the method of the embodiments of the present invention may be included in the application programs 4022.

[0130] In an embodiment of the present invention, by invoking the programs or instructions stored in the memory 402, specifically, the programs or instructions stored in the application program 4022, the processor 401 is used to execute the method steps provided in each method embodiment, for example, including:

[0131] When an object search request is obtained, retrieve the target department corresponding to the object search request from the knowledge base, where the knowledge base includes: the objectives and key results data of each department, and the objectives and key results data of the objects included in each department;

[0132] For each object in the target department, obtain the objectives and key results data from the knowledge base;

[0133] Perform a context reverse search operation according to the objectives and key results data of each object, so as to determine the target object from multiple objects through the context reverse search operation.

[0134] In a possible implementation manner, obtain the objectives and key results texts corresponding to multiple departments, and obtain the objectives and key results texts corresponding to the objects included in each department to obtain an objectives and key results set;

[0135] Perform invalid data filtering processing, technical entity recognition processing, and semantic disambiguation processing on the objectives and key results set;

[0136] Perform vector quantization encoding on the processed objectives and key results set to obtain the objectives and key results data of each department and the objectives and key results data of each object;

[0137] Generate the knowledge base with the objectives and key results data of each department and the objectives and key results data of each object.

[0138] In a possible implementation manner, when it is detected that any of the objectives and key results texts is updated, update the objectives and key results data in the knowledge base according to the update content.

[0139] In a possible implementation manner, when it is detected that any of the objectives and key results data in the knowledge base has not been updated for a preset duration, adjust the time decay factor of the unupdated objectives and key results data.

[0140] In a possible implementation manner, after performing semantic parsing on the object search request, perform intent recognition to obtain a recognition result;

[0141] Input the recognition result into a trained large model, so as to retrieve the target department corresponding to the recognition result from the knowledge base through the large model.

[0142] In a possible implementation, optimize the prompt words of the large model according to the object search request and the objective and key result data of all objects in the target department to obtain new prompt words;

[0143] So that the large model performs a context reverse search operation on the objects in the target department in the knowledge base according to the new prompt words to obtain the target object.

[0144] In a possible implementation, obtain explicit feedback and / or implicit feedback for the target object, where the explicit feedback includes: the satisfaction with the target object, and the implicit feedback includes: the referral behavior for the target object;

[0145] Optimize the large model according to the explicit feedback and / or the implicit feedback. The explicit feedback is used to optimize the weight configuration of the large model for the target object, and the implicit feedback is used to optimize the object recommendation strategy of the large model.

[0146] In a possible implementation, send a confirmation message to the target object;

[0147] When a response message returned by the target object for the confirmation message is not received within a preset duration, determine a new target object corresponding to the object search request.

[0148] The method disclosed in the embodiments of the present invention above can be applied to the processor 401 or implemented by the processor 401. The processor 401 may be an integrated circuit chip with signal processing capabilities. During implementation, the steps of the above method can be completed by the integrated logic circuit in hardware or instructions in software form in the processor 401. The above-mentioned processor 401 may be a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software units in the decoding processor. The software unit may be located in a mature storage medium in the art such as a random access memory, flash memory, read-only memory, programmable read-only memory, or electrically erasable programmable memory, register, etc. This storage medium is located in the memory 402, and the processor 401 reads the information in the memory 402 and combines its hardware to complete the steps of the above method.

[0149] It can be understood that these embodiments described herein can be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For a hardware implementation, the processing unit can be implemented in one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or a combination thereof.

[0150] For a software implementation, the technologies described herein can be implemented by units that execute the functions described herein. The software code can be stored in a memory and executed by a processor. The memory can be implemented inside or outside the processor.

[0151] The computer device provided in this embodiment may be the computer device as shown in Figure 7 and can execute all the steps of the method as shown in Figures 1-3 , thereby achieving the technical effects of the method as shown in Figures 1-3 . For specific reference, please refer to Figures 1-3 for the relevant description. For the sake of brevity, it will not be elaborated here.

[0152] The embodiment of the present invention also provides a storage medium (computer-readable storage medium). One or more programs are stored in the storage medium here. Among them, the storage medium may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, hard disk or solid-state drive; the memory may also include a combination of the above types of memory.

[0153] When one or more programs in the storage medium can be executed by one or more processors to implement the above object search method executed on the device side.

[0154] The processor is used to execute the object search program stored in the memory to implement the following steps of the object search method executed on the device side:

[0155] When an object search request is obtained, retrieve the target department corresponding to the object search request from the knowledge base, where the knowledge base includes: the goals and key result data of each department, and the goals and key result data of the objects included in each department;

[0156] For each object in the target department, obtain the goals and key result data from the knowledge base;

[0157] Perform a context reverse search operation according to the goals and key result data of each object to determine the target object from multiple objects through the context reverse search operation.

[0158] In a possible implementation manner, obtain the goals and key result texts corresponding to multiple departments, and obtain the goals and key result texts corresponding to the objects included in each department to obtain a goals and key result set;

[0159] Perform invalid data filtering processing, technical entity recognition processing, and semantic disambiguation processing on the goals and key result set;

[0160] Perform vector quantization encoding on the processed goals and key result set to obtain the goals and key result data of each department and the goals and key result data of each object;

[0161] Generate the knowledge base with the goals and key result data of each department and the goals and key result data of each object.

[0162] In a possible implementation, when it is detected that any of the said objectives and key results text is updated, the objectives and key results data in the knowledge base are updated according to the updated content.

[0163] In a possible implementation, when it is detected that any objectives and key results data in the knowledge base have not been updated for more than a preset duration, the time decay factor of the unupdated objectives and key results data is adjusted.

[0164] In a possible implementation, after semantic parsing of the object search request, intent recognition is performed to obtain a recognition result;

[0165] The recognition result is input into a trained large model to retrieve the target department corresponding to the recognition result from the knowledge base through the large model.

[0166] In a possible implementation, the prompt words of the large model are optimized according to the object search request and the objectives and key results data of all objects in the target department to obtain new prompt words;

[0167] So that the large model performs a context reverse search operation on the objects in the target department according to the new prompt words to obtain the target object.

[0168] In a possible implementation, explicit feedback and / or implicit feedback for the target object are obtained. The explicit feedback includes: satisfaction with the target object, and the implicit feedback includes: referral behavior for the target object;

[0169] The large model is optimized according to the explicit feedback and / or the implicit feedback. The explicit feedback is used to optimize the weight configuration of the large model for the target object, and the implicit feedback is used to optimize the object recommendation strategy of the large model.

[0170] In a possible implementation, a confirmation message is sent to the target object;

[0171] When a response message returned by the target object for the confirmation message is not received within a preset duration, a new target object corresponding to the object search request is determined.

[0172] Those skilled in the art should further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the present invention.

[0173] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0174] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. An object search method, characterized in that: include: When an object search request is obtained, the target department corresponding to the object search request is retrieved from a knowledge base, wherein the knowledge base includes: the target and key achievement data of each department, and the target and key achievement data of the objects contained in each department; For each object in the target department, obtaining target and key result data from the knowledge base; A contextual reverse lookup operation is performed based on the objectives and key achievement data of each of the objects, so as to determine a target object from the plurality of objects through the contextual reverse lookup operation.

2. The method according to claim 1, characterized in that Before obtaining the object search request, the method further includes: Obtaining the target and key result texts corresponding to the multiple departments, and obtaining the target and key result texts corresponding to the objects included in each department, to obtain a target and key result set; Perform invalid data filtering, technical entity identification and semantic disambiguation processing on the set of objectives and key results; Vectorize and encode the processed set of goals and key results to obtain the goals and key results data of each department and the goals and key results data of each object; The knowledge base is generated by combining the objectives and key achievement data of each department and the objectives and key achievement data of each object.

3. The method according to claim 2, characterized in that The method further comprises: When it is detected that any of the objectives and key results texts are updated, the objectives and key results data in the knowledge base are updated according to the updated content.

4. The method according to claim 1, characterized in that: The method further comprises: When it is detected that any target and key achievement data in the knowledge base has not been updated for more than a preset period of time, the time decay factor of the unupdated target and key achievement data is adjusted.

5. The method according to claim 1, characterized in that: The step of retrieving the target department corresponding to the object search request from the knowledge base includes: After semantically parsing the object search request, perform intent recognition to obtain a recognition result; The recognition result is input into the trained big model so as to retrieve the target department corresponding to the recognition result from the knowledge base through the big model.

6. The method according to claim 1, characterized in that The contextual reverse query operation based on the target and key result data of each object includes: Optimizing the prompt words of the large model according to the object search request and the goals and key results data of all objects in the target department to obtain new prompt words; The large model performs a context reverse lookup operation on the objects in the target department in the knowledge base according to the new prompt word to obtain the target object.

7. The method according to claim 1, characterized in that After determining the target object from the plurality of objects through the context reverse lookup operation, the method further includes: Obtaining explicit feedback and / or implicit feedback for the target object, wherein the explicit feedback includes: satisfaction for the target object, and the implicit feedback includes: forwarding recommendation behavior for the target object; The large model is optimized according to the explicit feedback and / or the implicit feedback, wherein the explicit feedback is used to optimize the weight configuration of the large model for the target object, and the implicit feedback is used to optimize the object recommendation strategy of the large model.

8. The method according to claim 1, characterized in that: After determining the target object from the plurality of objects through the context reverse lookup operation, the method further includes: Sending a confirmation message to the target object; When no response message returned by the target object to the confirmation message is received within a preset time period, a new target object corresponding to the object search request is determined.

9. An object search device, characterized in that: include: A retrieval module, for retrieving the target department corresponding to the object search request from a knowledge base when an object search request is obtained, wherein the knowledge base includes: the target and key achievement data of each department, and the target and key achievement data of the objects contained in each department; An acquisition module, for acquiring target and key result data from the knowledge base for each object in the target department; The determination module is used to perform a context reverse lookup operation based on the target and key achievement data of each of the objects, so as to determine a target object from the plurality of objects through the context reverse lookup operation.

10. A computer device, characterized in that: include: A processor and a memory, wherein the processor is used to execute an object search program stored in the memory to implement the object search method according to any one of claims 1 to 8.

11. A storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the object search method according to any one of claims 1 to 8.

Images