Dialogue analysis method, device and equipment for employment assistant

Abstract

The embodiment of the invention discloses a dialogue analysis method, device and equipment for an employment assistant. The method comprises the steps of obtaining an evaluation service type of the employment assistant; building a user agent based on the evaluation service type; the construction of the user agent comprises the steps of obtaining conflict-free features corresponding to the evaluation service type, constructing derivative features, generating a role template according to the conflict-free features and the derivative features, and performing reasoning according to the role template and historical dialogue information to generate user dialogue data; storing user dialogue data returned by the user agent and reply data of the employment assistant in response to the user dialogue data, and obtaining a dialogue record; and dialogue analysis of the employment assistant is completed according to the dialogue record.

Description

TECHNICAL FIELD

[0001] The present specification relates to the technical field of artificial intelligence, and in particular, to a dialogue analysis method, device and equipment of an employment assistant. BACKGROUND

[0002] Artificial intelligence (AI) technology breaks down the information barriers and service cost limitations between human resource companies and recruitment platforms by accurately matching supply and demand. In this context, to ensure the service quality of intelligent employment assistants, the field of automated dialogue analysis for intelligent employment assistants has emerged, which mainly simulates the behavior of job seekers based on a key component, User Agent, to evaluate the dialogue performance of employment assistants.

[0003] In the current dialogue analysis system for agents, the role characteristics of the current user agent are generated by a fixed feature combination set by artificial setting to generate a role template (Role), and the dialogue generation of the role is simulated based on a single prompt word template under the role template.

[0004] However, with the increase of application scenarios and dialogue complexity, generating a role template based on a fixed feature combination is prone to have a single dimension and logical conflicts, which further leads to a lack of realism in the dialogue role. With the increase of dialogue turns, the user agent only gives the model a single fixed prompt word at the beginning of the dialogue, and the single prompt word template is difficult to maintain the continuity of the role characteristics due to the influence of context information decay, making it difficult for the user agent to produce dialogue style and role characteristics coupling, and unable to guarantee the natural embodiment of role characteristics in long dialogue.

[0005] Therefore, there is a need for an employment assistant dialogue analysis method that can enhance the realism and continuity of role dialogue. SUMMARY

[0006] One or more embodiments of the present specification provide an employment assistant dialogue analysis method, device and equipment, which can solve the technical problem of the need for an employment assistant dialogue analysis method that can enhance the realism and continuity of role dialogue.

[0007] To solve the above technical problems, one or more embodiments of the present specification are implemented as follows: The employment assistant dialogue analysis method provided by one or more embodiments of the present specification comprises: obtaining an evaluation business type of the employment assistant; constructing a user intelligent agent based on the evaluation business type; the constructing of the user intelligent agent comprises: acquiring non-conflict features corresponding to the evaluation business type, constructing derivative features, generating a role template according to the non-conflict features and the derivative features, reasoning according to the role template and historical dialogue information, and generating user dialogue data; saving the user dialogue data returned by the user intelligent agent and reply data of the employment assistant in response to the user dialogue data, and obtaining a dialogue record; completing dialogue analysis of the employment assistant according to the dialogue record.

[0008] An embodiment of the present specification provides a dialogue analysis device of an employment assistant, comprising: an acquisition module configured to acquire an evaluation business type of an employment assistant; a construction module configured to construct a user intelligent agent based on the evaluation business type; the constructing of the user intelligent agent comprises: acquiring non-conflict features corresponding to the evaluation business type, constructing derivative features, generating a role template according to the non-conflict features and the derivative features, reasoning according to the role template and historical dialogue information, and generating user dialogue data; a dialogue module configured to save the user dialogue data returned by the user intelligent agent and reply data of the employment assistant in response to the user dialogue data, and obtain a dialogue record; an evaluation module configured to complete dialogue analysis of the employment assistant according to the dialogue record.

[0009] An embodiment of the present specification provides a dialogue analysis device of an employment assistant, comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: acquire an evaluation business type of an employment assistant; construct a user intelligent agent based on the evaluation business type; the constructing of the user intelligent agent comprises: acquiring non-conflict features corresponding to the evaluation business type, constructing derivative features, generating a role template according to the non-conflict features and the derivative features, reasoning according to the role template and historical dialogue information, and generating user dialogue data; save the user dialogue data returned by the user intelligent agent and reply data of the employment assistant in response to the user dialogue data, and obtain a dialogue record; complete dialogue analysis of the employment assistant according to the dialogue record.

[0010] The above at least one technical solution adopted by the embodiments of the present specification can achieve the following beneficial effects: The combination of the non-conflicting features and the derived features generates the role template, and the user dialogue data is generated based on the role template and the historical dialogue information, so that the user agent can simulate a virtual user with rich features and real behavior, ensuring consistent feature logic while overcoming the problem of single role caused by fixed prompt word templates, providing a more realistic test environment for the employment assistant. Save the complete dialogue record, and infer the historical dialogue information when generating user data dialogue, effectively maintaining the context relevance in the long dialogue process. Avoid the decoupling of dialogue style and role features caused by context information decay in traditional methods, improving the coherence and information consistency of long dialogue. Based on the evaluation business type of the employment assistant, the non-conflicting features are obtained and the derived features are constructed, so that the generated role template can accurately match the evaluation needs of different business types, realizing the dialogue analysis of the employment assistant. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present specification or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present specification, and those skilled in the art can also obtain other drawings according to these drawings without creative labor. In the drawings: Figure 1 A flowchart of a dialogue analysis method of an employment assistant provided by an embodiment of the present specification; FIG. 2(a) is an opening segment interaction example diagram of an employment assistant provided by one or more embodiments of the present specification; FIG. 2(b) is an interview segment interaction example diagram of an employment assistant provided by one or more embodiments of the present specification; FIG. 2(c) is a job recommendation segment interaction example diagram of an employment assistant provided by one or more embodiments of the present specification; Figure 3 For one or more embodiments of the present specification, a user portrait diagram in an application scenario is provided; Figure 4 For an embodiment of the present specification, a dialogue style diagram in an application scenario is provided; Figure 5 For an embodiment of the present specification, a user agent construction diagram in an application scenario is provided; Figure 6 For an embodiment of the present specification, a user role template diagram in an application scenario is provided; Figure 7 Provided in the embodiments of the present specification is a user agent dynamic simulation user sketch in an application scenario. Figure 8 Provided in the embodiments of the present specification is a structural schematic diagram of a dialogue analysis device of an employment assistant. Figure 9 Provided in the embodiments of the present specification is a structural schematic diagram of a dialogue analysis device of an employment assistant. DETAILED DESCRIPTION

[0012] The embodiments of the present specification provide a dialogue analysis method, device and equipment of an employment assistant.

[0013] In order for those skilled in the art to better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be described clearly and completely in conjunction with the drawings in the embodiments of the present specification. Obviously, the described embodiments are only part of the embodiments of the present specification, not all. Based on the embodiments of the present specification, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present specification.

[0014] In the scenario mentioned in the background, the current dialogue analysis method for the employment assistant is also the dialogue evaluation. With the expansion of the employment assistant from simple job query recommendation to complex interaction according to career planning, salary expectation and skill evaluation, the employment assistant accurately recommends personalized positions for users. When generating a job seeker role template with a fixed feature combination, not only will the job seekers be too similar due to the single feature dimension (such as only including basic attributes such as major and education), making it difficult to simulate real-world job-seeking scenarios. More importantly, there is no verified logical relationship between fixed features, which can cause self-contradictory problems in the role template. In the process of continuous multi-round dialogue of the employment assistant, with the continuous accumulation of dialogue context, the role feature of the role prompt word relying only on the initial word input will be affected by the attention mechanism dilution, which will destroy the authenticity and continuity of the dialogue process.

[0015] More intuitively, an example of a solution that the applicant has tried and the specific problems involved in the solution are introduced.

[0016] In a dialogue analysis method of an employment assistant, a user agent first completes the construction of the required features of the user template based on the Cartesian product combination of limited features such as age, education and dialect. At the same time, this version realizes the basic multi-round dialogue function through a prompt word calling a large language model.

[0017] Although this solution realizes user simulation for the employment assistant, the user agent sends messages to the employment assistant end through simulated dialogues to obtain the reply data of the employment assistant for evaluation. However, the user agent relies on the Cartesian product combination of limited features, resulting in a single feature dimension that is difficult to expand and lacks comprehensiveness. Moreover, there are frequent logical conflicts between features, such as the unrealistic role templates of "19-year-old with 10 years of work experience" or "54-year-old graduate." Moreover, the role features are not naturally reflected in the dialogues. In the single-agent mode, the long prompt words cause the context information to gradually decay, and the closer to the end of the dialogue, the greater the impact of the information on the large language model. The generated dialogue frequently contains random codes or bracket residues, affecting the clarity and coherence of the dialogue. There is also a dialogue history amnesia problem, and in multi-round dialogues, there may be inconsistencies between the information provided before and after.

[0018] Figure 1 For one or more embodiments of the present specification, a flowchart of a dialogue analysis method of an employment assistant is provided. The method can be applied to different business evaluation fields, such as public employment, university career guidance, commercial recruitment platform, human resource outsourcing services, and intelligent evaluation fields such as travel service evaluation and financial service evaluation in addition to employment-related fields. The flowchart can be executed by a computing device corresponding to the relevant field (such as a server corresponding to public employment, etc.). Some input parameters or intermediate results in the flowchart allow for manual intervention to adjust to help improve accuracy.

[0019] Figure 1 The flowchart in the present specification can include the following steps: S102: Obtain the evaluation business type of the employment assistant.

[0020] As shown in FIG. 2(a)-2(c), the employment assistant is programmed by the pointer code, deployed on the client, based on the API interface or other forms, and achieves the service of deep mining of employment intention and accurate recommendation of positions through three-stage multi-round dialogue. Among them, in combination with FIG. 2(a)-2(c), the three-stage multi-round dialogue includes: an opening section of guiding the user to enter the interaction process and establishing a preliminary trust relationship; a user interview section of collecting the core information of the user's professional background, skill preference and salary expectation through structured questions; and a job recommendation section of dynamically generating personalized job recommendations and providing a feedback mechanism based on the user portrait and the job database. The job recommendation section preferentially interfaces with the positions having a relevant cooperation relationship with the business assistant, and judges whether to complete the recommendation or continue to recommend new positions according to the user's click operation, that is, the user uploaded interaction content. For example: if the user is not interested in the recommended position, the employment assistant will analyze the reason and determine whether to persuade or recommend a new position. If the user clicks the one-click registration button, the registration is completed, and the standard reply content corresponding to the button is returned. In addition, the job recommendation section can also provide a job card for public users to jump to the job detail page.

[0021] The evaluation business type, that is, which core capabilities or interaction scenarios of the employment assistant need to be evaluated and analyzed, so as to determine the characteristic range of the user intelligent body corresponding to the evaluation business type and the dimension to be evaluated. Further, the evaluation business type includes but is not limited to: resume optimization consultation, interview skill guidance, position recommendation matching, salary and welfare negotiation, etc. Each evaluation business type corresponds to a corresponding role template and a targeted service process. For example, if the evaluation business type is salary and welfare negotiation, the subsequent user intelligent body will simulate a user entering the salary negotiation stage.

[0022] Specifically, in one possible embodiment, the evaluation business type can be obtained based on the following manner. First, the evaluation system or the evaluation service end administrator or the business party receives the business type selection instruction submitted through the preconfigured interface, and determines the evaluation business type required by the employment assistant according to the business type selection instruction. The preconfigured interface can provide interactive components such as drop-down menus, radio buttons or text input boxes for the administrator or the business party to select or input the business type identifier.

[0023] In addition, in another possible embodiment, the evaluation business type can also be obtained in the following manner: according to the preconfigured evaluation plan, the preset evaluation business types are loaded in sequence. For example, a complete dialogue analysis plan sequentially stipulates to evaluate the three business types of "position recommendation matching", "resume optimization consultation" and "interview skill guidance", and each evaluation business type can be loaded in sequence.

[0024] S104: constructing a user agent based on the evaluation business type; the construction of the user agent includes: obtaining a conflict-free feature corresponding to the evaluation business type, constructing a derived feature, generating a role template according to the conflict-free feature and the derived feature, reasoning according to the role template and historical dialogue information, and generating user dialogue data.

[0025] In the current agent-oriented employment assistant analysis mode, the role features of the user agent depend on the fixed feature combination set by artificial, and the role template is generated as Figure 3 The user portrait role module, and the dialogue generation of the role under the single prompt word template. However, the generation of the role template based on the fixed feature combination is prone to problems such as single dimension and logical conflict. Moreover, with the increase of dialogue turns, the single prompt word template is difficult to maintain the continuity of the role features in long dialogue, so that the user agent generates dialogue style as Figure 4 indicated, and Figure 3 The role features of the role template are difficult to couple, and it is difficult to ensure the natural embodiment of the user role features in long dialogue.

[0026] Therefore, as shown in Figure 5 , by obtaining the conflict-free feature corresponding to the evaluation business type, the problem of generating the role module by relying on the fixed feature combination set by artificial for the role features of the user agent is solved, and the problems of single dimension and logical conflict are prone to occur. Then, based on the conflict-free feature, the derived feature is constructed, and the role template is generated according to the combination of the conflict-free feature and the derived feature. By adding the derived feature, the limitation of the fixed feature combination is solved, and the authenticity of the evaluation scene is improved. When generating each round of user dialogue, the role template and historical dialogue information are input into the thinking chain engine for reasoning, which is equivalent to re-injecting the role of the user agent in each round of dialogue, effectively avoiding the problem that the role features are difficult to continuously penetrate in long dialogue due to information decay.

[0027] Specifically, in the current user agent construction method, when defining the role features, whether relying on the limited feature combination preset by artificial or randomly extracting features from the database, there may be logical conflict problems between the features. Therefore, to solve the problems of single dimension and logical conflict of the role features, in one or more embodiments of the present specification, the conflict-free feature corresponding to the evaluation business type is obtained, and the derived feature is constructed, which specifically includes the following processes: The multi-dimensional role characteristics corresponding to the evaluation business type are obtained. For example, the characteristic dimensions and their candidate values associated with the evaluation business type are called from a pre-stored role characteristic library according to the evaluation business type. The role characteristic library is a structured database that stores various characteristics of job seekers. Optionally, the multi-role characteristic dimensions include, but are not limited to, basic identity characteristics including education, work experience, and age, professional ability characteristics including job positions, professional skills, and project experience, job intention characteristics including expected salary ranges, target industries, preferred company sizes, work cities, and job urgency, and behavior style characteristics including communication styles, role inclinations, and negotiation styles. For example, the evaluation business type in a certain application scenario is salary and welfare negotiation, and under this evaluation business type, characteristics strongly associated with negotiation scenarios such as “current salary”, “expected salary increase”, “other offer situation in hand”, and “salary negotiation experience” are obtained. Under the evaluation business type of resume optimization consulting, corresponding role characteristics such as “work experience”, “project results”, “skill certificates”, and “resume writing ability” are obtained.

[0028] To combine the non-conflicting characteristics in the above-mentioned role characteristics, non-conflicting characteristics of multi-dimensional role characteristics need to be identified to combine non-conflicting characteristics to generate user role characteristics. Since the role characteristics obtained above corresponding to the evaluation business type are randomly obtained and combined, it may produce roles that violate common sense, resulting in subsequent dialogue being untrustworthy. Therefore, conflict identification and filtering of role characteristics are needed, such as matching and filtering through a basic conflict rule library to obtain non-conflicting characteristics, to generate user role characteristics by Cartesian product combination of non-conflicting characteristics as a stable basic role setup.

[0029] After obtaining the user role characteristics, to achieve reasonableness while improving the diversity of the role, the user role characteristics are expanded through a large language model to obtain derivative characteristics. Through the semantic understanding and generation capability of the large language model in this process, the problem of single role dimension and stereotyped behavior is solved, and the realism of subsequent user agent dialogue is improved.

[0030] Specifically, in the process of obtaining user role characteristics by combining non-conflicting candidate characteristics through Cartesian product, Cartesian product generates all possible characteristic combinations, and filtering based only on the basic conflict rule library is difficult to filter role characteristic combinations that have no hard conflicts but have low correlation and do not exist in actual applications. Therefore, to further filter out invalid roles and improve the effectiveness of subsequent dialogue, in one or more embodiments of the present specification, combining non-conflicting characteristics to generate user role characteristics specifically includes: By identifying the logical conflicts of the multi-dimensional role features, the non-conflict candidate features are obtained. The identification of the logical conflicts can be achieved by inputting the obtained multi-dimensional role features into a preset basic conflict rule library for checking. Based on the above-mentioned obtained non-conflict candidate features, the correlation degree between each of the non-conflict candidate features is determined, and the non-conflict candidate features are identified again to obtain non-conflict features. In one feasible embodiment, the secondary identification process is a vectorization process on the non-conflict candidate features, such as inputting the description text of each non-conflict candidate feature into the embedding layer of the pre-trained language model, thereby converting it into a role feature vector in a high-dimensional space, so that the semantics of the non-conflict candidate features are retained in the spatial position and direction of the role feature vector. The semantic correlation degree between the role feature vectors is determined by the cosine similarity or dot product between them. It can be understood that the similarity between the "fresh graduate" vector and the "30-40k" vector is lower than the correlation degree between the "fresh graduate" vector and the "5-10k" vector. When the semantic correlation degree is determined to be lower than a certain threshold, the role feature combination is marked with a conflict flag. Collect all features that do not trigger the conflict flag and determine them as non-conflict features. After obtaining the non-conflict features with high correlation degree through two levels of screening based on the above steps, the non-conflict features are enumerated, and all enumerated feature combinations, such as Cartesian product combination or dot product combination, are combined into a structured user role feature.

[0031] Specifically, in order to integrate the derived features obtained after semantic conflict identification and deep expansion into a complete role template, in one or more embodiments of the present specification, the role template is generated according to the non-conflict features and the derived features, including the following steps: After combining and packaging the non-conflict features and the derived features according to the preset structured format, the candidate role template is obtained. The non-conflict features constitute the basic attribute framework of the role, that is, the Figure 6 The specific features include fixed features and professional features, wherein the fixed features include but are not limited to name, gender, age, work experience, smoke prevention, city of residence, education, etc., and the professional features include but are not limited to work history, expected salary, and health certificate. In addition, based on the description of the above embodiment, it can be known that the derived features are obtained after the expansion of the non-conflict features, such as Figure 6The abstract features of the role template shown include feature data such as job-seeking state, interview attitude, expression ability, and intention clarity generated by a large language model. For example: based on the non-conflict features (education: "bachelor's degree", work experience: "2 years", job position: "Java development engineer", current salary: "10K-13K") and derived features ("understanding of Spring framework stays at the basic use level", "actively ask for hints when encountering difficult problems in interviews", "prefer teams with technical mentor training system"). Through the combination of the above non-conflict features and derived features, a candidate role template of "looking for a junior Java engineer for growth" can be generated. In order to ensure that the generated candidate role template meets the logic and predetermined template requirements, and ensure that there is no logical conflict or inappropriate features, the candidate role will be audited based on the pre-set logic constraints, and multiple role templates that pass the audit will be obtained. The audit process can be implemented based on the Self_Check module. Then, in order to use different role templates to simulate conversations with the employment assistant in different evaluation rounds, after obtaining the role templates, each role template is labeled and stored in the pre-set template library, so that based on the label of each role template, the corresponding role template is called from the pre-set template library in turn.

[0032] In the specific implementation process of generating user dialogue data based on role templates and historical dialogue information, in order to balance the evaluation quality and efficiency, in one feasible embodiment provided in the present embodiment, the role template corresponding to the current evaluation round and the dialogue examples associated with the corresponding role template are obtained.

[0033] The role template corresponding to the current evaluation round is obtained by determining the role template corresponding to the current evaluation round according to the evaluation scale, the label of the role template, and the current evaluation round. That is, the evaluation scale of the employment assistant is obtained, and it is determined whether the evaluation scale is full evaluation or partial evaluation. It can be understood that, based on the evaluation demand of the current employment assistant, it is determined whether to select the full evaluation evaluation scale or the partial evaluation evaluation scale. Full evaluation is suitable for employment assistant deep evaluation that needs to cover all role scenarios completely, and partial evaluation is suitable for representative scenario verification under resource limited conditions. If the current evaluation mode is full evaluation, each template is traversed in turn according to the label of the role template to determine the role template corresponding to each evaluation round, so that each role template is called in the corresponding evaluation round. If it is partial evaluation, the role templates corresponding to each evaluation round are obtained by proportionally sampling from the preset template library based on uniform sampling. For example, assuming that there are 1000 role templates and the evaluation scale is full evaluation, and 1000 rounds of evaluation are generated, then the role template with a label of 1 is used for the first round of evaluation. If the evaluation scale is partial evaluation, and 500 rounds of evaluation are generated, then 500 labels are randomly sampled, the role template corresponding to the first label after random sampling is used for the first round of evaluation, and so on.

[0034] The dialog example associated with the corresponding role template is obtained, that is, the standard dialog record associated with the determined role template is retrieved from the example library to provide a scene reference for subsequent dialog generation.

[0035] After obtaining the role template corresponding to the current round based on the above steps, the relevant historical dialog information is called according to the similarity between the current reply data of the employment assistant and the historical dialog information, so as to maintain the coherence of the dialog logic. The similarity between the current reply data of the employment service and the historical dialog information can be calculated by converting the current reply data and the historical dialog information into vector representations through a semantic encoding model such as Sentence-BERT, and then calculating the cosine similarity to construct an association mapping. The relevant historical dialog information is filtered through a similarity threshold to exclude redundant, irrelevant, or covered by subsequent dialog historical dialog information. After obtaining the relevant historical dialog information, because the original dialog information usually contains redundant information, directly inputting the thinking chain engine will occupy the context window and distract the model attention, so as to improve the inference efficiency, the relevant historical dialog information is summarized to obtain a key historical dialog summary. To generate user dialog data that meets the role setting of the role template and maintains the coherence of the dialog to respond to the interactive information of the employment assistant, the corresponding role template, dialog example, and key historical dialog summary are input into the preset thinking chain engine for reasoning to generate user dialog data.

[0036] The preset chain-of-thought (COT) engine is a technical method in the field of artificial intelligence and natural language processing, which can simulate the human thinking process of solving problems by making the model generate a series of intermediate reasoning steps, and decompose the complex dialogue generation task into multiple logically related thinking steps to generate user dialogue data through the thinking steps.

[0037] In one or more embodiments of the present specification, the user agent needs to generate user dialogue data according to the user role template and historical dialogue information. In order to improve the controllability of the generation process and the quality of the generated results, the present embodiment also includes generating user dialogue data based on a preset workflow framework. The preset workflow framework is a Workflow module, which functions to decompose complex tasks into multiple modules and schedule and execute them in a process-oriented manner. Through the design of the preset workflow framework, that is, the Workflow module, the scalability and maintainability of the dialogue can be improved.

[0038] Based on the preset workflow framework, the generation of user dialogue data includes generating a dialogue generation task of a user agent based on a role template and historical dialogue information. The historical dialogue information refers to the dialogue record and its context information of the user dialogue data generated by the user agent simulating the role template and the reply data of the employment assistant in the current evaluation round, that is, the current session. In one possible embodiment, the generation process of the dialogue generation task is to perform structured analysis on the role template to extract key feature fields therein. Optionally, the key feature fields include at least one of basic identity features, professional skill features, behavior style features, and job-seeking intention features. After analyzing the role template data, the historical dialogue information needs to be analyzed to obtain the context information. For example, the historical dialogue information is subjected to semantic recognition to obtain an analysis result of the historical dialogue information. Based on the analyzed role template data and the analysis result of the historical dialogue information, the key parameters of task generation are determined. After determining the task generation parameters, the relevant data needs to be encapsulated into a standard task data structure to instantiate the task data structure to obtain the dialogue generation task.

[0039] After generating the dialogue generation task, because dialogue generation involves multiple dimensional requirements, a single generation model cannot meet the multi-dimensional requirements, so the dialogue generation task is horizontally decomposed to obtain multiple task execution units. That is, the dialogue generation task is decomposed into multiple independent sub-tasks according to the functional dimension to obtain the task execution unit. Optionally, the task execution unit in a certain application scenario can include at least one of an intent analysis unit, a role consistency verification unit, a context connection unit, a language stylization unit, and a content security filtering unit.

[0040] To ensure that each task execution unit can accurately perform its specific function, appropriate prompt words need to be assigned to each unit. The prompt word templates corresponding to each task execution unit are obtained from the pre-set prompt word template library, and the prompt words are generated in combination with the output results of the upstream task execution units. For example, for the role consistency verification unit, its prompt word template may include placeholders for role occupation background, role personality characteristics, etc. At this time, the specific role template information of the upstream task unit output result will be filled into these placeholders to obtain the prompt word corresponding to the current task execution unit. After obtaining the prompt words of each task execution unit, based on the execution order of each task execution unit defined by the pre-set workflow framework, each task execution unit is executed in turn, and each task execution unit performs specific processing based on its corresponding prompt word, and transmits the processing result to the downstream task execution unit to generate initial user dialogue data. Since there may be abnormal data in the model generation process, causing garbled characters or bracket residue problems in the initial user dialogue data. After generating the initial user dialogue data, the garbled data of the user dialogue data is identified, and the garbled data is filtered to obtain the user dialogue data. The garbled data refers to data containing meaningless characters, logical confusion, and syntax errors, etc. Quality problems can be identified by a pre-set character rule setting encoding program or based on natural language processing. Through the cleaning of garbled data, the clarity and coherence of the user dialogue data are improved.

[0041] Based on the above, the user agent can simulate different types of users through different role templates, produce distinctive dialogue styles, and interact with the employment assistant. As shown in Figure 7 The user agent can simulate an 18-year-old graduate or a 46-year-old mother to realize dialogue interaction between the user agent simulating a user role and the employment assistant. However, in this process, the dialogue system corresponding to the current user agent mostly only supports text interaction, lacks scalable voice and multi-modal functions, and is difficult to cover more complex application scenarios. In order to improve the application scenario adaptability of the user agent, make it can realize natural interaction from text to voice based on role templates and employment assistants, and improve the authenticity of the dialogue agent. In one or more embodiments of the present specification, the user dialogue data returned by the user agent is input into the employment assistant to obtain reply data, which can be realized through the following process: First, according to the dialogue scenario of the employment assistant, the output format of the user dialogue data is determined. Among them, the dialogue scenario type is divided into text dialogue scenario and voice dialogue scenario. The text dialogue scenario refers to the interaction between the user agent and the employment assistant through text, and the voice dialogue scenario refers to the interaction between the user agent and the employment assistant through voice. The dialogue scenario of the employment assistant can be determined by analyzing the endpoint configuration and protocol support information of the employment assistant. After identifying the dialogue scenario type, the output format of the user dialogue data is determined according to the dialogue scenario type. According to the output format requirement, the corresponding data conversion channel is selected, and the related processing module is initialized. Among them, in order to realize the multi-modal interaction as shown in Figure 7 Fig. 1, if it is a text channel corresponding to a text dialogue type, the text data can be directly transmitted without special conversion. If it is a voice channel corresponding to a voice dialogue type, the text returned by the user agent needs to be converted into voice data by a text-to-speech (TTS) module and transmitted to the employment assistant, so that the employment assistant can broadcast the user dialogue data through voice to obtain the reply data of the employment assistant in response to the user dialogue data. Similarly, if the reply data of the employment assistant is based on a voice dialogue scenario, the input voice signal needs to be converted into processable text information through automatic speech recognition (ASR) to realize the mapping function from voice to text.

[0042] The user dialogue data returned by the user agent is input into the employment assistant, and the interface response diagram of the employment assistant in Fig. 2 (a)-Fig. 2 (c) and Figure 7 Fig. 2 (d) can know that the employment assistant will generate reply data corresponding to the user dialogue data. In one or more embodiments of the present specification, the employment assistant generates reply data corresponding to the user dialogue data by the following way: The employment assistant will identify the business link of the employment assistant according to the user dialogue data and the last reply data in the dialogue record of the current evaluation round. According to the business link, the reply logic of the employment assistant is matched, so as to process the user dialogue data based on the reply logic and generate the current reply data. Among them, the business link of the employment assistant includes opening speech link, user interview link and job recommendation link as shown in Fig. 2 (a)-Fig. 2 (c).

[0043] According to the business link, the reply logic of the employment assistant is matched, so that the user dialogue data is processed based on the reply logic to generate the current reply data. For different business links, if the current is in the opening link, the trigger instruction in the user dialogue data, that is, the specific button click event, is analyzed to automatically call and execute the preset interview process corresponding to the quality to generate the process entry information as the reply data. For example: based on the intelligent agent simulating clicking the waiter, a response window connection is popped up to enter the AI interview link, that is, the current process entry information, so as to enter the next business link of the employment assistant. If it is in the user interview link, the user dialogue data is matched with the pre-set structured question tree, when the user dialogue data meets the trigger condition of a specific question, the next structured question in the question tree is automatically taken as the reply data, so as to collect the structured information of the user, facilitate the employment assistant to build the user portrait for subsequent job recommendation. If it is in the job recommendation link, the user dialogue data is subjected to intent recognition and key information extraction, the employment database is matched based on the extraction result, and personalized recommendation content is generated as the reply data according to the feedback information obtained by the matching. By returning the reply data to the user intelligent agent, the dialogue of the current evaluation round can continue.

[0044] In another feasible embodiment, to solve the problem that the role model cannot simulate the emotional fluctuations of real users in the job seeking process due to the lack of emotional state change mechanism, resulting in rigid dialogue response mode and insufficient emotional continuity in multi-round dialogue. The method further comprises: based on the evaluation business type, constructing a plurality of user intelligent agents. Through the construction of multiple intelligent agents, the multi-level modeling of the role and the cooperation mechanism between the intelligent agents are introduced, realizing the simulation of the user intelligent agent for the complex real job seeking scene, and achieving the service evaluation of the employment assistant under high concurrency and multi-style user requests.

[0045] Among them, the construction of multiple user intelligent agents includes the following processes: The evaluation dimensions of the evaluation business type are obtained to generate a static role script according to the non-conflict features and derived features corresponding to each evaluation dimension. Among them, the evaluation dimensions of the employment assistant in a certain application scenario include: positive multi-round dialogue, employment induction inquiry, employment intent diversity, and non-employment intent dialogue. Different evaluation dimensions can be set based on the evaluation requirements of different evaluation business types of the employment assistant. After obtaining the evaluation dimensions, the non-conflict features corresponding to each evaluation dimension can be determined according to the core feature types of each evaluation dimension, and the derived features are obtained based on the extension mode of the non-conflict features in the above embodiment. The non-conflict features and derived features are combined, and the static role script is established after being checked and verified.

[0046] After constructing the static role script, it is necessary to extract emotional dynamic information from historical dialogues. In one possible embodiment, by analyzing historical dialogue information, obtain emotion state transition data to generate dynamic role scripts according to static role scripts and emotion state transition data. That is, by preprocessing the historical dialogue information through natural language processing technology, obtain the processed key interaction sequence. The preprocessing includes word segmentation, entity recognition, intent extraction and emotion labeling. The key interaction sequence is the user dialogue data that can trigger the change of emotional state and the corresponding job assistant reply data. According to the key interaction sequence, the emotion state evolution data is constructed, which is used to record the initial user emotion, trigger event, target user emotion and emotion duration conversation times. Based on the emotion state evolution data, the transition condition probability between any two emotional states is calculated, that is, the probability of converting from emotion state 1 to emotion state 2 when a given event type is triggered. For example: when the trigger event is "interview failure feedback", the transition probability of each emotional state is: calm to anxious, probability 0.75; worried to anxious, probability 0.82; expect to worry, probability 0.68. When the trigger event is "offer notification", the transition probability of each emotional state is: anxious to joy, probability 0.95; worried to joy, probability 0.91. Organize these probability values into a state transition probability matrix, where the row represents the initial emotional state, the column represents the target emotional state, and the matrix element is the transition condition probability. The state transition matrix is the emotion state transition data used to quantify the emotional change rule of the user agent.

[0047] The behavior decision weight is used to define the influence of the emotional state and the characteristics of the static role script in the intelligent dialogue behavior. Based on the correlation between the characteristics of the static role script and the current job assistant evaluation dimension, the basic weight is determined, and the basic weight is dynamically corrected based on the emotion transition probability in the state transition probability matrix, such as multiplying the basic weight by the relevant emotion transition probability and the preset adjustment coefficient to obtain the behavior decision weight. Based on the static role script as the basic framework, the behavior decision weight is used to search the preset emotion behavior database to match the corresponding emotion expression sentence, tone characteristics and dialogue content tendency to obtain the dynamic role emotion. In order to realize the collaborative interaction of multiple user agents and achieve the purpose of simulating the high-concurrency scene of multiple users requesting services at the same time in a real scene, according to the shared data pool corresponding to the multiple user agents and the dynamic role script corresponding to each user agent, the user dialogue data is generated by reasoning and sent to the job assistant in parallel. That is, each user agent generates dialogue data that meets its own emotional and demand state by reasoning through a reinforcement learning model based on its own dynamic role script, combined with public information in the shared data pool and the interaction state of other agents.

[0048] S106: Save the user dialogue data returned by the user agent and the reply data of the employment assistant in response to the user dialogue data, and obtain a dialogue record.

[0049] To solve the problem of lack of effective memory mechanism in multi-round dialogue process, resulting in the user agent being unable to maintain the coherence of the dialogue, and problems such as contradictions and repeated inquiries. As shown in Figure 7 The user agent simulates a role of an 18-year-old graduate or a 46-year-old mother by establishing a role template, and the result of each round of dialogue between the user agent and the employment assistant when the user agent simulates the user role and interacts with the employment assistant is recorded as Figure 7 dialogue memory and as Figure 5 multi-round dialogue record, so as to be used for subsequent evaluation work. In addition, the dialogue record is also memorized by the user agent, and accurate memory information is provided in subsequent dialogue.

[0050] S108: Complete the dialogue analysis of the employment assistant according to the dialogue record.

[0051] To realize whether the employment assistant has poor service effect and abnormal service quality, and to ensure the continuous improvement of the employment evaluation service quality, it is necessary to analyze the dialogue of the employment assistant according to the dialogue record after the user agent simulates the user dialogue with the user template and interacts with the employment assistant. The dialogue analysis process of the employment assistant focuses on evaluating the performance of the employment assistant in response to the user agent simulating different user dialogue styles, including indicators such as fluency, accuracy, emotional resonance, and coherence of multi-round interaction.

[0052] In a feasible embodiment, the dialogue analysis of the employment assistant is completed according to the dialogue record, specifically including: The dialog records of each evaluation round are extracted from the persistent storage dialog records to construct the evaluation set of the employment assistant. After constructing the evaluation set, the multi-dimensional indicators of the evaluation set need to be quantitatively evaluated based on the pre-set calculation logic and pre-set evaluation standards to obtain the analysis results of each evaluation sample. Among them, the dimensions of the employment assistant that need to be analyzed include: employment induction inquiry, employment intention diversity, and non-employment intention dialogue from four dimensions. And the specific multi-dimensional indicators include but are not limited to: evaluation set size, overall efficiency, AI interview oral, AI interview task compliance ability, AI interview understanding, AI interview memory, AI interview content authenticity, AI interview task completion, AI job recommendation oral AI job recommendation task compliance ability, AI job recommendation understanding, AI job recommendation memory, AI job recommendation content authenticity, AI job recommendation task completion. Then, based on the conversation to which each evaluation sample belongs, the analysis results of each evaluation sample are aggregated to obtain single-round analysis results, such as: the analysis results of all evaluation samples belonging to the same evaluation round are statistically aggregated, the average score and fluctuation of each dialogue in the evaluation round are calculated, the change trend of each indicator in the evaluation round is analyzed, the key performance nodes and abnormal situations in the evaluation round are identified, and the single-round analysis results containing dialogue feature analysis are generated. After obtaining the single-round analysis results, the single-round analysis results need to be aggregated to realize global aggregation analysis to obtain the analysis results of each dimension indicator, so as to compare the different indicator performances of different business scenarios such as AI interview and AI job recommendation. Through the multi-index aggregation analysis of the employment assistant, the comprehensive evaluation of the employment assistant ability is realized, which provides data support for the optimization and ability improvement of the employment assistant.

[0053] In addition to analyzing the dialog of the employment assistant, the performance of the user agent can also be analyzed, and the performance analysis results will help further optimize the performance of the user agent to maintain efficient and natural conversation ability in different situations. As shown in Table 1, 3456 Role templates are constructed for a certain application scenario, and the multi-dimensional evaluation indicator results of the dialog analysis of the employment assistant from the four dimensions of forward multi-round dialogue, employment induction inquiry, employment intention diversity, and non-employment intention dialogue are shown. Table 2 is an evaluation indicator description table of the user agent in a certain application scenario. The performance of the user agent can be analyzed based on the indicator description in Table 2, and the service of the user agent can be optimized according to the performance analysis results.

[0054] Table 1. Multi-dimensional evaluation indicator result table

[0055] Table 2. Evaluation indicator description table of user agent

[0056] Based on the same idea, one or more embodiments of the present specification also provide a device and equipment corresponding to the above method, as shown in Figure 8 、 Figure 9 .

[0057] Figure 8 A structural schematic diagram of a dialogue analysis device of an employment assistant provided by one or more embodiments of the present specification, the device comprising: An acquisition module 802 acquires an evaluation business type of an employment assistant; A construction module 804 constructs a user agent based on the evaluation business type; the construction of the user agent includes: acquiring non-conflict features corresponding to the evaluation business type, constructing derived features, generating a role template according to the non-conflict features and the derived features, reasoning according to the role template and historical dialogue information, and generating user dialogue data; A dialogue module 806 saves the user dialogue data returned by the user agent and the reply data of the employment assistant in response to the user dialogue data, and obtains a dialogue record; An evaluation module 808 completes dialogue analysis of the employment assistant according to the dialogue record.

[0058] Optionally, the acquisition module 802 acquires multi-dimensional role features corresponding to the evaluation business type; Identify the non-conflict features of the multi-dimensional role features, and combine the non-conflict features to generate user role features; Expand the user role features through a large language model to obtain derived features.

[0059] Optionally, the acquisition module 802 identifies the logical conflicts of the multi-dimensional role features to obtain non-conflict candidate features; Determine the correlation degree between each of the non-conflict candidate features, and perform secondary identification on the non-conflict candidate features to obtain non-conflict features; Enumerate and combine the non-conflict features to generate user role features.

[0060] Optionally, the construction module 804 combines the non-conflict features and the derived features to obtain a candidate role template; Based on a pre-set logical constraint condition, the candidate role is audited to obtain a plurality of role templates that pass the audit; Wherein, after obtaining the role template, each of the role templates will be labeled and stored in a pre-set template library.

[0061] Optionally, the construction module 804 acquires a role template corresponding to a current evaluation round and a dialogue example associated with the corresponding role template; According to the similarity between the current reply data of the employment assistant and the historical dialogue information, relevant historical dialogue information is called; By summarizing the relevant historical dialogue information, a key historical dialogue summary is obtained; The corresponding role template, the dialogue example and the key historical dialogue summary are input into a preset thinking chain engine for reasoning to generate user dialogue data.

[0062] Optionally, the construction module 804 acquires an evaluation scale of the employment assistant; According to the evaluation scale, the label of the role template and the current evaluation round, a role template corresponding to the current evaluation round is determined.

[0063] Optionally, the device further comprises a dialogue acquisition module 810; The dialogue acquisition module 810 is based on a preset workflow framework to generate the user dialogue data; The user dialogue data includes generating a dialogue generation task of the user agent based on the role template and the historical dialogue information; The dialogue generation task is horizontally decomposed to obtain a plurality of task execution units; According to the prompt word template corresponding to the task execution unit and the output result of the upstream task execution unit, a corresponding prompt word is allocated to it; Based on the corresponding prompt word, the execution of each task execution unit is guided to generate initial user dialogue data; The user dialogue data is identified and the garbage data is filtered to obtain user dialogue data.

[0064] Optionally, the dialogue module 806 determines the output format of the user dialogue data according to the dialogue scenario of the employment assistant; Through a data conversion channel corresponding to the output format, the user dialogue data returned by the user agent is input into the employment assistant to obtain reply data.

[0065] Optionally, the dialogue module 806 identifies the business link of the employment assistant based on the user dialogue data and the last reply data of the employment assistant; According to the business link, the reply logic of the employment assistant is matched to process the user dialogue data based on the reply logic to generate current reply data.

[0066] Optionally, the evaluation module 808 constructs an evaluation set of the employment assistant by taking the dialogue records as evaluation samples; According to the preset evaluation standard, the multi-dimensional indicators of the evaluation set are quantitatively evaluated to obtain analysis results of each evaluation sample; Based on the conversation to which each evaluation sample belongs, the analysis results of each evaluation sample are aggregated to obtain single-round analysis results, and the single-round analysis results are aggregated to obtain analysis results of each dimension indicator.

[0067] Optionally, the device further comprises a multi-agent construction module 812; The multi-agent construction module 812 constructs a plurality of user agents based on the evaluation business type; wherein the construction of the plurality of user agents comprises: Obtaining evaluation dimensions of the evaluation business type to generate static role scripts according to non-conflict features and derived features corresponding to each evaluation dimension; Obtaining emotion state transition data from the historical dialogue information to generate dynamic role scripts according to the static role scripts and the emotion state transition data; According to the shared data pool corresponding to the plurality of user agents and the dynamic role scripts corresponding to each user agent, user dialogue data is generated by reasoning and sent to the employment assistant in parallel.

[0068] Optionally, the multi-agent construction module 812 identifies a key interaction sequence of the historical dialogue information, and determines emotion state evolution data according to the key interaction sequence; According to the state transition condition probability corresponding to the emotion state evolution data, a state transition probability matrix is constructed to generate emotion state transition data; Combining each state transition condition probability in the emotion state transition data with the features of the static role script, a behavior decision weight is obtained; According to the behavior decision weight and the static role script, a dynamic role script is integrated to obtain; wherein the dynamic role script can find a preset emotion behavior database based on the behavior decision weight to obtain a dynamic role emotion.

[0069] Figure 9 A structural schematic diagram of a dialogue analysis device of an employment assistant is provided for one or more embodiments of the present specification, and the device comprises: At least one processor; and, The memory is in communication connection with the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Obtain an evaluation business type of the employment assistant; Based on the evaluation business type, a user agent is constructed; the construction of the user agent includes: obtaining a conflict-free feature corresponding to the evaluation business type, constructing a derived feature, generating a role template according to the conflict-free feature and the derived feature, reasoning according to the role template and historical dialogue information, and generating user dialogue data; Save the user dialogue data returned by the user agent and the reply data of the employment assistant in response to the user dialogue data to obtain a dialogue record; According to the dialogue record, the dialogue analysis of the employment assistant is completed.

[0070] In the 1990s, it was possible to distinguish whether an improvement in a technology was a hardware improvement (e.g., an improvement in the circuit structure of a diode, transistor, switch, etc.) or a software improvement (an improvement in a method flow). However, as technology has advanced, many improvements in method flows today can be considered as direct improvements in hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into a hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is an integrated circuit whose logic function is determined by user programming of the device. A designer programs a digital system "integrated" on a PLD by himself / herself, without having to ask a chip manufacturer to design and manufacture a special integrated circuit chip. Moreover, instead of manually manufacturing an integrated circuit chip, this programming is now mostly implemented using "logic compiler" software, which is similar to a software compiler used when developing a program, and the original code before compilation must also be written in a specific programming language, which is called a hardware description language (HDL), and there are many types of HDL, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., and the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that it is only necessary to logically program a method flow using the above-mentioned hardware description languages and program it into an integrated circuit to easily obtain a hardware circuit that implements the logical method flow.

[0071] The controller can be implemented in any suitable way, for example, the controller can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code, such as software or firmware, executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that, in addition to implementing the controller purely in computer readable program code, it is possible to implement the same functionality with logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. The controller can thus be considered a hardware component, and the means included therein for implementing the various functions can be considered structures within the hardware component. Alternatively, or even additionally, the means for implementing the various functions can be considered both software modules implementing the method and structures within the hardware component.

[0072] The systems, apparatuses, modules or units illustrated by the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0073] For the sake of description, the above apparatuses are described in various units by functions respectively. Of course, the functions of each unit can be implemented in one or more software and / or hardware in implementing the present specification.

[0074] Those skilled in the art will understand that the embodiments of the present specification can be provided as a method, a system or a computer program product. Therefore, the embodiments of the present specification can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present specification can take the form of a computer program product implemented 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.

[0075] The specification is presented with reference to flow diagrams and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams 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 processing element 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, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow or multiple flows and / or blocks Figure 1 one or more flow or multiple flows and / or blocks

[0076] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow or multiple flows and / or blocks Figure 1 one or more flow or multiple flows and / or blocks

[0077] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more flow or multiple flows and / or blocks Figure 1 one or more flow or multiple flows and / or blocks

[0078] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0079] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory, etc. The memory is an example of computer readable media.

[0080] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0081] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0082] The specification can be described in the general context of computer-executable instructions, such as program modules, executed by computers. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The specification can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0083] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for device, equipment, non-volatile computer storage medium embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.

[0084] The above-described embodiments of the application have special structure and can achieve the desired results. Other embodiments can have different structures and achieve the same results. The purpose of the above-described embodiments is to illustrate the principles of the application and not to limit the scope of the application. The scope of the application is defined by the claims and their equivalents. Other embodiments are within the scope of the claims.

[0085] The above description is merely illustrative of the embodiments of the present application and is not intended to limit the scope of the present application. Various modifications can be made by those skilled in the art based upon the teachings disclosed herein. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present application shall fall within the scope of the claims of the present application.

Claims

1. A dialogue analysis method for an employment assistant, the method comprising: Obtain the assessment service type from the employment assistant; Based on the aforementioned evaluation service type, a user intelligent agent is constructed; The construction of the user intelligent agent includes: acquiring conflict-free features corresponding to the evaluation service type, constructing derived features, generating a role template based on the conflict-free features and the derived features, and performing inference based on the role template and historical dialogue information to generate user dialogue data. Save the user dialogue data returned by the user intelligent agent and the employment assistant's response data in response to the user dialogue data to obtain the dialogue record; Based on the dialogue records, complete the dialogue analysis of the employment assistant.

2. The method as described in claim 1, wherein obtaining conflict-free features corresponding to the evaluation service type and constructing derived features specifically includes: Obtain multi-dimensional role characteristics corresponding to the evaluation business type; Identify the conflict-free features of the multi-dimensional role characteristics, and combine the conflict-free features to generate user role characteristics; The user role features are extended using a large language model to obtain derived features.

3. The method as described in claim 2, wherein identifying the conflict-free features of the multi-dimensional role characteristics and combining the conflict-free features to generate user role characteristics specifically includes: Identify logical conflicts in the multi-dimensional role features and obtain conflict-free candidate features; Determine the correlation degree between each of the conflict-free candidate features, and perform secondary identification on the conflict-free candidate features to obtain conflict-free features; The conflict-free features are enumerated and combined to generate user role features.

4. The method as described in claim 1, wherein generating a character template based on the conflict-free features and the derived features includes: The conflict-free features and the derived features are combined to obtain candidate role templates; Based on pre-set logical constraints, the candidate roles are reviewed to obtain multiple role templates that have passed the review. After obtaining the character templates, each character template will be marked and stored in a preset template library.

5. The method as described in claim 1, wherein reasoning is performed based on the role template and historical dialogue information to generate user dialogue data, specifically including: Obtain the character template corresponding to the current evaluation round, and the dialogue example associated with the corresponding character template; Based on the similarity between the current response data of the employment assistant and the historical dialogue information, relevant historical dialogue information is retrieved; By summarizing the relevant historical dialogue information, a summary of key historical dialogues is obtained; The corresponding role template, the dialogue example, and the key historical dialogue summary are input into a pre-built thought chain engine for reasoning to generate user dialogue data.

6. The method described in claim 5, obtaining the character template corresponding to the current evaluation round, specifically includes: Obtain the assessment scale of the employment assistant; Based on the evaluation scale, the character template markings, and the current evaluation round, determine the character template corresponding to the current evaluation round.

7. The method of claim 5, further comprising: The generation of user dialogue data is based on a pre-built workflow framework; The generation of user dialogue data includes: generating a dialogue generation task for the user agent based on the role template and the historical dialogue information; The dialogue generation task is horizontally decomposed to obtain multiple task execution units; Based on the prompt word template corresponding to the task execution unit and the output result of the upstream task execution unit, a corresponding prompt word is assigned to it; Based on the corresponding prompt words, the execution of each task execution unit is guided to generate initial user dialogue data; Identify and filter the garbled data in the user dialogue data to obtain the user dialogue data.

8. The method as described in claim 1, wherein the user dialogue data returned by the user intelligent agent is input into the employment assistant to obtain response data, specifically includes: Based on the dialogue scenario of the employment assistant, determine the output format of the user dialogue data; The user dialogue data returned by the user agent is input into the employment assistant to obtain response data through a data conversion channel corresponding to the output format.

9. The method as described in claim 8, wherein the user dialogue data returned by the user intelligent agent is input into the employment assistant to obtain response data, specifically includes: Based on the user dialogue data and the previous reply data of the employment assistant, the business stage in which the employment assistant is located is identified; Based on the aforementioned business process, the employment assistant's response logic is matched to process the user dialogue data and generate the current response data.

10. The method as described in claim 1, wherein the dialogue analysis of the employment assistant is performed based on the dialogue records, specifically including: The dialogue records were used as evaluation samples to construct the evaluation set for the employment assistant; Based on the pre-set calculation logic and pre-set evaluation criteria, the multi-dimensional indicators of the evaluation set are quantitatively evaluated to obtain the analysis results of each evaluation sample. Based on the session to which each evaluation sample belongs, the analysis results of each evaluation sample are aggregated to obtain the single-round analysis results, and the single-round analysis results are aggregated to obtain the analysis results of each dimension index.

11. The method of claim 1, further comprising: Based on the aforementioned evaluation service type, multiple user intelligent agents are constructed; wherein, the construction of the multiple user intelligent agents includes: Obtain the evaluation dimensions of the evaluation business type, and generate static character scripts based on the conflict-free features and derived features corresponding to each evaluation dimension; By using the historical dialogue information, emotional state transition data is obtained, and a dynamic character script is generated based on the static character script and the emotional state transition data. Based on the shared data pool corresponding to the multiple user agents and the dynamic role scripts corresponding to each user agent, user dialogue data is inferred and generated, and sent to the employment assistant in parallel.

12. The method as described in claim 11, wherein emotional state transition data is obtained through the historical dialogue information, and a dynamic character script is generated based on the static character script and the emotional state transition data, specifically including: Identify key interaction sequences in the historical dialogue information, and determine emotional state evolution data based on the key interaction sequences; Based on the state transition conditional probabilities corresponding to the emotional state evolution data, a state transition probability matrix is ​​constructed to generate emotional state transition data. By combining the conditional probabilities of each state transition in the emotional state transition data with the characteristics of the static character script, the behavioral decision weights are obtained. Based on the behavioral decision weights and the static character script, a dynamic character script is obtained by integration; wherein, the dynamic character script can obtain dynamic character emotions by searching a pre-set emotional behavior database based on the behavioral decision weights.

13. A dialogue analysis device for an employment assistant, comprising: The module retrieves the assessment service types of the employment assistant. The construction module constructs the user intelligent agent based on the evaluation service type; The construction of the user intelligent agent includes: acquiring conflict-free features corresponding to the evaluation service type, constructing derived features, generating a role template based on the conflict-free features and the derived features, and performing inference based on the role template and historical dialogue information to generate user dialogue data. The dialogue module saves the user dialogue data returned by the user intelligent agent and the employment assistant's response data in response to the user dialogue data, and obtains the dialogue record; The evaluation module performs dialogue analysis on the employment assistant based on the dialogue records.

14. The apparatus of claim 13, wherein the acquisition module acquires multi-dimensional role characteristics corresponding to the evaluation service type; Identify the conflict-free features of the multi-dimensional role characteristics, and combine the conflict-free features to generate user role characteristics; The user role features are extended using a large language model to obtain derived features.

15. The apparatus of claim 14, wherein the acquisition module identifies logical conflicts in the multi-dimensional role features and obtains conflict-free candidate features; Determine the correlation degree between each of the conflict-free candidate features, and perform secondary identification on the conflict-free candidate features to obtain conflict-free features; The conflict-free features are enumerated and combined to generate user role features.

16. The apparatus of claim 13, wherein the construction module combines the conflict-free feature and the derived feature to obtain a candidate role template; Based on pre-set logical constraints, the candidate roles are reviewed to obtain multiple role templates that have passed the review. in, After obtaining the character templates, each character template will be marked and stored in a preset template library.

17. The apparatus of claim 13, wherein the construction module acquires a character template corresponding to the current evaluation round and a dialogue example associated with the corresponding character template; Based on the similarity between the current response data of the employment assistant and the historical dialogue information, relevant historical dialogue information is retrieved; By summarizing the relevant historical dialogue information, a summary of key historical dialogues is obtained; The corresponding role template, the dialogue example, and the key historical dialogue summary are input into a pre-built thought chain engine for reasoning to generate user dialogue data.

18. The apparatus of claim 17, wherein the construction module obtains the assessment scale of the employment assistant; Based on the evaluation scale, the character template markings, and the current evaluation round, determine the character template corresponding to the current evaluation round.

19. The apparatus of claim 17, further comprising: Dialogue acquisition module; The dialogue acquisition module generates user dialogue data based on a pre-built workflow framework. The generation of user dialogue data includes: generating a dialogue generation task for the user agent based on the role template and the historical dialogue information; The dialogue generation task is horizontally decomposed to obtain multiple task execution units; Based on the prompt word template corresponding to the task execution unit and the output result of the upstream task execution unit, a corresponding prompt word is assigned to it; Based on the corresponding prompt words, the execution of each task execution unit is guided to generate initial user dialogue data; Identify and filter the garbled data in the user dialogue data to obtain the user dialogue data.

20. The apparatus of claim 13, wherein the dialogue module determines the output format of the user dialogue data based on the dialogue scenario of the employment assistant; The user dialogue data returned by the user agent is input into the employment assistant to obtain response data through a data conversion channel corresponding to the output format.

21. The apparatus of claim 20, wherein the dialogue module identifies the business stage of the employment assistant based on the user dialogue data and the employment assistant's previous response data; Based on the aforementioned business process, the employment assistant's response logic is matched to process the user dialogue data and generate the current response data.

22. The apparatus of claim 13, wherein the evaluation module uses the dialogue records as evaluation samples to construct an evaluation set for the employment assistant; Based on the pre-set calculation logic and pre-set evaluation criteria, the multi-dimensional indicators of the evaluation set are quantitatively evaluated to obtain the analysis results of each evaluation sample. Based on the session to which each evaluation sample belongs, the analysis results of each evaluation sample are aggregated to obtain the single-round analysis results, and the single-round analysis results are aggregated to obtain the analysis results of each dimension index.

23. The apparatus of claim 13, further comprising: Multi-agent building blocks; The multi-agent construction module constructs multiple user agents based on the evaluation service type; wherein the construction of the multiple user agents includes: Obtain the evaluation dimensions of the evaluation business type, and generate static character scripts based on the conflict-free features and derived features corresponding to each evaluation dimension; By using the historical dialogue information, emotional state transition data is obtained, and a dynamic character script is generated based on the static character script and the emotional state transition data. Based on the shared data pool corresponding to the multiple user agents and the dynamic role scripts corresponding to each user agent, user dialogue data is inferred and generated, and sent to the employment assistant in parallel.

24. The apparatus of claim 23, wherein the multi-agent construction module identifies key interaction sequences in the historical dialogue information and determines emotional state evolution data based on the key interaction sequences; Based on the state transition conditional probabilities corresponding to the emotional state evolution data, a state transition probability matrix is ​​constructed to generate emotional state transition data. By combining the conditional probabilities of each state transition in the emotional state transition data with the characteristics of the static character script, the behavioral decision weights are obtained. Based on the behavioral decision weights and the static character script, a dynamic character script is obtained by integration; wherein, the dynamic character script can obtain dynamic character emotions by searching a pre-set emotional behavior database based on the behavioral decision weights.

25. A dialogue analysis device for an employment assistant, comprising: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Obtain the assessment service type from the employment assistant; Based on the evaluation service type, a user intelligent agent is constructed; the construction of the user intelligent agent includes: acquiring conflict-free features corresponding to the evaluation service type, constructing derived features, generating a role template based on the conflict-free features and the derived features, and inferring user dialogue data based on the role template and historical dialogue information. Save the user dialogue data returned by the user intelligent agent and the employment assistant's response data in response to the user dialogue data to obtain the dialogue record; Based on the dialogue records, complete the dialogue analysis of the employment assistant.

Images