Interview interaction data processing method and system based on knowledge guidance

Abstract

The invention provides an interview interaction data processing method and system based on knowledge guidance, and the method comprises the steps: obtaining the current context content in each conversation of a current interview interaction process, carrying out the splicing processing based on the current context content and an action mapping strategy, and obtaining a first prompt word; inputting the first cue word into a language model, and obtaining at least one target action strategy adopted in the output next round of dialogue; processing the obtained interview guide information and at least one target action strategy based on a language model to obtain an interview interaction problem of a next round of conversation output by the language model; and under the condition of determining that the current interview interaction process is completed based on the language model, obtaining all output dialogue interaction contents corresponding to the current interview interaction process, and analyzing all the dialogue interaction contents to obtain a dialogue interaction content analysis result. According to the method provided by the invention, the interview interaction data processing efficiency can be effectively improved, so that the labor cost is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for processing interview interaction data based on knowledge guidance, and also to an electronic device and a non-transitory computer-readable storage medium. Background Art

[0002] In recent years, with the continuous development of artificial intelligence technology, various smart devices have been used more and more widely in people's daily interview scenarios. Interview is a widely used method with far-reaching influence in the field of qualitative research. The core concept of structured interview is to ensure that each interview uses exactly the same questions and asks them in the same order. This standardization ensures that answers can be reliably summarized and can be fully compared between different subgroups of the sample or between different survey periods. On the basis of structured interviews, semi-structured interviews break the limitations of fixed question sets and predefined orders, and ask exploratory questions about the details that arise during the interview process, thereby revealing deeper knowledge and deeper connections while maintaining sample comparability similar to structured interviews. However, conducting semi-structured interviews requires extensive participation from experienced users, which severely limits the efficiency of data collection and processing. In order to automate this process, the technical solution adopted by the existing technology is the Task-Oriented Dialogue (TOD), which is designed to respond to user input within a predefined action space. By parsing natural language expressions into specific ontologies, the dialogue system then tracks the state and selects an action to generate a response that meets the expected function. However, since the semi-structured interview process is usually controlled by the interviewer and conducted in the form of questions and answers, it aims to obtain information from the interviewee. However, the interview process based on the dialogue system is very time-consuming and requires the interviewer to have considerable experience, which greatly limits the efficiency and feasibility of data collection and processing. Therefore, how to design a more efficient knowledge-guided interview interaction data processing solution that can be applied to the semi-structured interview process has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a method for processing interview interaction data based on knowledge guidance, so as to solve the defect that the interview interaction data processing scheme in the prior art has high limitations and leads to poor actual efficiency.

[0004] The present invention provides a method for processing interview interaction data based on knowledge guidance, comprising: In each round of dialogue in the current interview interaction process, current context content is obtained, and splicing processing is performed based on the current context content and a preset action mapping strategy to obtain a first prompt word; Calling a preset language model, inputting the first prompt word into the language model, and obtaining at least one target action strategy to be adopted in the next round of dialogue output by the language model; processing the acquired interview guide information and the at least one target action strategy based on the language model, and obtaining the interview interaction question for the next round of dialogue output by the language model; When it is determined based on the language model that the current interview interaction process is completed, obtaining all the dialogue interaction contents output by the language model and corresponding to the current interview interaction process; The entire dialogue interaction content includes at least one round of interview interaction questions generated by the dialogue and the historical interaction content input by the interviewee; The language model is called to analyze all the dialogue interaction contents to obtain a dialogue interaction content analysis result.

[0005] According to the method for processing interview interaction data based on knowledge guidance of the present invention, before obtaining the current context content in each round of dialogue in the current interview interaction process, the method further includes: Obtaining a target question input by an interviewee, performing splicing processing based on the target question and a preset question mapping strategy to obtain a second prompt word; calling a preset language model, inputting the second prompt word into the language model, and obtaining at least one interrogation question associated with the target question output by the language model; wherein the question mapping strategy includes a plurality of question mapping relationships corresponding to each target question; and the interrogation question is a branch question or an extended question based on the target question; Historical conversation data including the target question and the at least one probing question is obtained, corresponding interview guide information is obtained based on the historical conversation data including the target question and the at least one probing question, and the interview guide information is stored in a database.

[0006] According to the interview interaction data processing method based on knowledge guidance of the present invention, the splicing process is performed based on the target question and the preset question mapping strategy to obtain a second prompt word; calling a preset language model, inputting the second prompt word into the language model, and obtaining at least one inquiry question associated with the target question output by the language model, specifically including: Performing a concatenation process based on the target question and the first question mapping relationship in the question mapping strategy to obtain a third prompt word; calling a preset language model, inputting the third prompt word into the language model, and obtaining a first analysis and processing result; Perform concatenation processing based on the target question and the second question mapping relationship in the question mapping strategy to obtain a fourth prompt word; call a preset language model, input the fourth prompt word into the language model, and obtain a second analysis processing result; Based on the target question and the third question mapping relationship in the question mapping strategy, a splicing process is performed to obtain a fifth prompt word; a preset language model is called, and the fifth prompt word is input into the language model to obtain at least one inquiry question associated with the target question; wherein the second prompt word includes the third prompt word, the fourth prompt word and the fifth prompt word; and the question mapping relationship includes a first question mapping relationship, a second question mapping relationship and a third question mapping relationship.

[0007] According to the knowledge-guided interview interaction data processing method of the present invention, before processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction questions for the next round of dialogue output by the language model, it also includes: acquiring the interview guide information corresponding to the current interview interaction process from the database.

[0008] According to the method for processing interview interaction data based on knowledge guidance of the present invention, the acquired interview guide information and the at least one target action strategy are processed based on the language model to obtain the interview interaction questions for the next round of dialogue output by the language model, including: The language model is called to combine the at least one target action strategy, and corresponding inquiry questions are generated based on the interview guide information to obtain interview interaction questions for the next round of dialogue output by the language model; wherein the at least one target action strategy includes at least one of a questioning action strategy, a responding action strategy, an inquiry action strategy, and a promotion action strategy.

[0009] According to the method for processing interview interaction data based on knowledge guidance of the present invention, when determining that the current interview interaction process is completed based on the language model, obtaining all the dialogue interaction contents output by the language model corresponding to the current interview interaction process specifically includes: Detecting an actual conversation turn of the current interview interaction process, and determining that the current interview interaction process has been completed when the actual conversation turn reaches a preset conversation turn threshold; After determining that the current interview interaction process has been completed, the dialogue interaction content of the current interview interaction process is summarized based on the language model to obtain all the dialogue interaction content corresponding to the current interview interaction process output by the language model.

[0010] The present invention also provides an interview interaction data processing system based on knowledge guidance, comprising: A prompt word acquisition module is used to obtain the current context content in each round of dialogue in the current interview interaction process, and perform splicing processing based on the current context content and a preset action mapping strategy to obtain a first prompt word; An interview interaction question acquisition module is used to call a preset language model, input the first prompt word into the language model, and obtain at least one target action strategy used in the next round of dialogue output by the language model; based on the language model, the acquired interview guide information and the at least one target action strategy are processed to obtain the interview interaction question for the next round of dialogue output by the language model; A dialogue interaction content acquisition module is used to obtain all dialogue interaction contents corresponding to the current interview interaction process output by the language model when it is determined that the current interview interaction process is completed based on the language model; wherein the all dialogue interaction contents include interview interaction questions generated by at least one round of dialogue and historical interaction contents input by the interviewee; The analysis module is used to call the language model to analyze all the dialogue interaction contents to obtain the dialogue interaction content analysis results.

[0011] According to the knowledge-guided interview interaction data processing system of the present invention, before obtaining the current context content in each round of dialogue in the current interview interaction process, the system further includes: Guide building blocks for: Obtaining a target question input by an interviewee, performing splicing processing based on the target question and a preset question mapping strategy to obtain a second prompt word; calling a preset language model, inputting the second prompt word into the language model, and obtaining at least one interrogation question associated with the target question output by the language model; wherein the question mapping strategy includes a plurality of question mapping relationships corresponding to each target question; and the interrogation question is a branch question or an extended question based on the target question; Historical conversation data including the target question and the at least one probing question is obtained, corresponding interview guide information is obtained based on the historical conversation data including the target question and the at least one probing question, and the interview guide information is stored in a database.

[0012] According to the knowledge-guided interview interaction data processing system of the present invention, the target question and the preset question mapping strategy are spliced ​​together to obtain a second prompt word; a preset language model is called, the second prompt word is input into the language model, and at least one inquiry question associated with the target question output by the language model is obtained, specifically including: Performing a concatenation process based on the target question and the first question mapping relationship in the question mapping strategy to obtain a third prompt word; calling a preset language model, inputting the third prompt word into the language model, and obtaining a first analysis and processing result; Perform concatenation processing based on the target question and the second question mapping relationship in the question mapping strategy to obtain a fourth prompt word; call a preset language model, input the fourth prompt word into the language model, and obtain a second analysis processing result; Based on the target question and the third question mapping relationship in the question mapping strategy, a splicing process is performed to obtain a fifth prompt word; a preset language model is called, and the fifth prompt word is input into the language model to obtain at least one inquiry question associated with the target question; wherein the second prompt word includes the third prompt word, the fourth prompt word and the fifth prompt word; and the question mapping relationship includes a first question mapping relationship, a second question mapping relationship and a third question mapping relationship.

[0013] According to the knowledge-guided interview interaction data processing system of the present invention, before processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction questions for the next round of dialogue output by the language model, it also includes: an interview guide acquisition module, which is used to acquire the interview guide information corresponding to the current interview interaction process from the database.

[0014] According to the knowledge-guided interview interaction data processing system of the present invention, the interview interaction question obtaining module is specifically used for: The language model is called to combine the at least one target action strategy, and corresponding inquiry questions are generated based on the interview guide information to obtain interview interaction questions for the next round of dialogue output by the language model; wherein the at least one target action strategy includes at least one of a questioning action strategy, a responding action strategy, an inquiry action strategy, and a promotion action strategy.

[0015] According to the knowledge-guided interview interaction data processing system of the present invention, the dialogue interaction content acquisition module is specifically used for: Detecting an actual conversation turn of the current interview interaction process, and determining that the current interview interaction process has been completed when the actual conversation turn reaches a preset conversation turn threshold; After determining that the current interview interaction process has been completed, the dialogue interaction content of the current interview interaction process is summarized based on the language model to obtain all the dialogue interaction content corresponding to the current interview interaction process output by the language model.

[0016] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for processing interview interaction data based on knowledge guidance as described in any one of the above items is implemented.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for processing interview interaction data based on knowledge guidance as described in any of the above items is implemented.

[0018] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the knowledge-guided interview interaction data processing method as described in any of the above items.

[0019] The present invention provides a method for processing interview interaction data based on knowledge guidance, which obtains the current context content in each round of dialogue in the current interview interaction process, performs splicing processing based on the current context content and the action mapping strategy, and obtains a first prompt word; the first prompt word is input into a language model to obtain at least one target action strategy adopted in the next round of dialogue output; the obtained interview guide information and at least one target action strategy are processed based on the language model to obtain the interview interaction questions of the next round of dialogue output by the language model; when it is determined based on the language model that the current interview interaction process is completed, all the dialogue interaction contents corresponding to the current interview interaction process are obtained, and all the dialogue interaction contents are analyzed to obtain the dialogue interaction content analysis results; the method can effectively improve the efficiency of interview interaction data processing, thereby greatly reducing the manpower cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 It is a flow chart of the interview interaction data processing method based on knowledge guidance provided by the present invention.

[0022] Figure 2 It is an interactive schematic diagram corresponding to the guide construction module provided by the prior art.

[0023] Figure 3 It is an interaction schematic diagram corresponding to the dialogue module provided by the present invention.

[0024] Figure 4It is a schematic diagram of system calls in the hardware environment provided by the present invention.

[0025] Figure 5 It is a schematic diagram of the operation of the knowledge-guided interview interaction data processing system provided by the present invention.

[0026] Figure 6 It is a structural schematic diagram of the knowledge-guided interview interaction data processing system provided by the present invention.

[0027] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Combine the following Figure 1-Figure 7 The present invention describes the interview interaction data processing method and system based on knowledge guidance, and describes its embodiments in detail.

[0030] The following is a detailed description of an embodiment of the method for processing interview interaction data based on knowledge guidance according to the present invention. Figure 1 As shown, it is a flow chart of the interview interaction data processing method based on knowledge guidance provided by the present invention, and the specific implementation process includes the following steps: Step 101: obtain the current context content in each round of dialogue in the current interview interaction process, perform splicing processing based on the current context content and a preset action mapping strategy, and obtain a first prompt word.

[0031] In an embodiment of the present invention, before obtaining the current context content in each round of dialogue in the current interview interaction process, an interview guide needs to be pre-built. Specifically, first obtain the target questions (i.e., the main questions of the interview) input by multiple interviewees, perform splicing processing based on the target questions and the preset question mapping strategy, and obtain the second prompt word; call the preset language model, input the second prompt word into the language model, and obtain at least one inquiry question associated with the target question output by the language model; wherein the question mapping strategy includes multiple question mapping relationships corresponding to each target question; the inquiry question is a branch question or an extended question associated with the target question that is inquired on the basis of the target question. Wherein, the language model is a preset large language model (LargeLanguage Model, LLM) called, which refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. Furthermore, historical conversation data including the target question and the at least one inquiry question may be obtained, and the historical conversation data may also include other conversation data in addition to the target question and the inquiry question. Corresponding interview guide information is obtained based on the historical conversation data including the target question and the at least one inquiry question (that is, the historical conversation data including the target question and the at least one inquiry question may be packaged into an interview guide to obtain interview guide information), and the interview guide information or interview guide is stored in a database.

[0032] In the process of performing splicing processing based on the target question and the preset question mapping strategy to obtain the second prompt word; calling the preset language model, inputting the second prompt word into the language model, and obtaining at least one inquiry question associated with the target question output by the language model, splicing processing can be performed based on the target question and the first question mapping relationship in the question mapping strategy to obtain the third prompt word; calling the preset language model, inputting the third prompt word into the language model, and obtaining the first analysis and processing result. Splicing processing is performed based on the target question and the second question mapping relationship in the question mapping strategy to obtain the fourth prompt word; calling the preset language model, inputting the fourth prompt word into the language model, and obtaining the second analysis and processing result. Splicing processing is performed based on the target question and the third question mapping relationship in the question mapping strategy to obtain the fifth prompt word; calling the preset language model, inputting the fifth prompt word into the language model, and obtaining at least one inquiry question associated with the target question; wherein the second prompt word includes the third prompt word, the fourth prompt word, and the fifth prompt word; and the question mapping relationship includes the first question mapping relationship, the second question mapping relationship, and the third question mapping relationship. Specifically, the first question mapping relationship may be a mapping relationship between the target question and the prompt word of the first step; the second question mapping relationship may be a mapping relationship between the target question and the prompt word of the second step; the third question mapping relationship may be a mapping relationship between the target question and the prompt word of the third step. In addition, the question mapping relationship may also include a fourth question mapping relationship. The fourth question mapping relationship may be a mapping relationship between the target question and the prompt word of the fourth step. Figure 2 As shown, the prompt word for the first step may be "understand the main problem or understand the main problem", and the "understand the main problem" is to let the knowledge model explain the meaning and function of each target problem, so that the subsequent output of the model is more specific and relevant; the prompt word for the second step may be "predict the direction of the conversation", and the "predict the direction of the conversation" is to let the knowledge model predict various situations that may occur in the interview based on the main problem; the prompt word for the third step may be "generate interrogative questions", and the "generate interrogative questions" is to let the knowledge model refer to the possible direction of the interview and design multiple interrogative questions that should be raised in different situations; the prompt word for the fourth step may be "propose quantitative indicators", and the "propose quantitative indicators" is to let the knowledge model output the corresponding quantitative indicators for each interviewee, and the proposed quantitative indicators are used in the analysis and processing process after the interview.

[0033] It should be noted that although the interviewer will make adjustments based on the actual progress of the semi-structured interview, such adjustments still need to be made within the predefined question framework or at least around the predefined question framework. This predefined question framework is called an interview guide, and the interview guide contains interview guide information. Therefore, the interview guide plays a key role in semi-structured interviews. The interview guide should contain two layers of open-ended questions: (1) main questions. These target questions involve broad topics of interest and are used to guide the overall direction of the interview conversation. Therefore, an interview guide needs to be constructed in advance; (2) probing questions. These probing questions are derived from the main questions and are designed to explore in depth specific points that are particularly valuable in the discussion. They are generated by the guide construction module. For example: enter a series of main questions {Mi}, the guide construction module (

[0034] Guide Construction Module (GCM) generates multiple query questions {Pi, j} for each of the main questions. That is:

[0035] Among them, GCM is a mapping from a main problem Mi to corresponding multiple query problems Pi,j.

[0036] The interview guide consists of the input main questions and the generated inquiry questions:

[0037] This generation involves bridging two differences between the two layers of questions. (1) General vs. Specific: Main questions establish the framework of the interview, while probing questions must delve into the details of each main question, which requires a thorough understanding of them; (2) Expected vs. Actual: Main questions outline the expected interview questions, while probing questions must cover potential valuable points that arise during the interview, which requires anticipation of the actual process. The present invention starts with the main questions provided, such as Figure 2 As shown. Specifically, in multiple rounds of dialogue, the knowledge model is guided to (1) understand the main issues to solve the first difference (i.e., general and specific), (2) predict the potential direction of the interview to solve the second difference (i.e., expected and actual), and finally (3) call the knowledge model to generate at least one inquiry question for each main issue. In addition, the present invention also designs an additional step, namely (4) indicator design, calling the knowledge model to generate quantitative indicators for each interviewee for subsequent analysis of the interview.

[0038] like Figure 4As shown in the figure, in the specific process of interview guide construction or generation: (1) through client (front-end) interaction, guide the user to input the main questions involved in the interview (i.e., target questions) in a given way. (2) Combined with the main questions, use the template designed in advance through prompt engineering to generate prompt words containing four steps to guide the large language model to achieve the following functions: understand the main questions: let the large language model explain the meaning and function of each main question, so that the subsequent output of the model is more specific and relevant; predict the direction of the conversation: let the large language model predict various situations that may occur in the interview based on the main questions; generate inquiry questions: let the large language model refer to the possible direction of the interview and design inquiry questions that should be raised in different situations; propose quantitative indicators: let the large language model generate quantitative indicators for each interviewee for subsequent analysis of the interview. (3) Integrate the output of the large language model in the above steps to obtain the interview guide.

[0039] Before executing this step, interview guide information corresponding to the current interview interaction process may be obtained from the database.

[0040] Step 102: call a preset language model, input the first prompt word into the language model, and obtain at least one target action strategy to be adopted in the next round of dialogue output by the language model; process the acquired interview guide information and the at least one target action strategy based on the language model, and obtain the interview interaction questions for the next round of dialogue output by the language model.

[0041] In an embodiment of the present invention, the language model is called to combine the at least one target action strategy, and corresponding inquiry questions are generated based on the interview guide information to obtain interview interaction questions for the next round of dialogue output by the language model; wherein the at least one target action strategy includes at least one of a questioning action strategy, a response action strategy, an inquiry action strategy, and a promotion action strategy.

[0042] The questioning action strategy may be a conversation content of the nature of asking questions to the interviewee, the response action strategy may be a conversation content of the nature of feeding back answers to the interviewee, the inquiry action strategy may be a conversation content of the nature of sending inquiring questions to the interviewee, and the advancement action strategy may be a conversation content of the nature of sending questions to the interviewee to advance the next step.

[0043] The question action strategy is Figure 3 The action strategy for the question action marked in blue in Figure 3 The blue-marked “Thank you for participating in the interview. Do you often use the big model?” output by the knowledge model (i.e., the agent) on the left is one of the dialogue contents corresponding to the question action strategy; the response action strategy is Figure 3 The action strategy for the response action marked in red in Figure 3 The red output of the knowledge model (i.e., the agent) on the left is “Indeed, big models play an increasingly important role in our lives”, which is one of the dialogue contents corresponding to the response action strategy; the inquiry action strategy is Figure 3 The action strategy of the inquiry action marked in orange in Figure 3 The orange output of the knowledge model (i.e., the agent) on the left is “Can you tell me some specific scenarios where you use the big model?”, which is one of the dialogue contents corresponding to the inquiry action strategy; the advancement action strategy is Figure 3 The purple-marked action strategy for the push action Figure 3 The purple-marked output of the knowledge model (i.e., the agent) on the left is "I just mentioned that you read books and wrote papers. Now, let's discuss some topics related to education." This is one of the dialogue contents corresponding to the action promotion strategy.

[0044] Step 103: When it is determined based on the language model that the current interview interaction process is completed, obtain all dialogue interaction content corresponding to the current interview interaction process output by the language model; wherein, all dialogue interaction content includes interview interaction questions generated by at least one round of dialogue and historical interaction content input by the interviewees.

[0045] In an embodiment of the present invention, the actual dialogue rounds of the current interview interaction process are detected, and when the actual dialogue rounds reach a preset dialogue round threshold (such as 10 dialogue rounds), it is determined that the current interview interaction process has been completed. After determining that the current interview interaction process has been completed, the dialogue interaction content of the current interview interaction process is summarized based on the language model to obtain all the dialogue interaction content corresponding to the current interview interaction process output by the language model.

[0046] It should be noted that in the processing of the dialogue module, the main advantage of semi-structured interviews is that they allow the interviewer to focus on the planned route while still giving the interviewer autonomy to explore relevant ideas that arise in the interview. However, these advantages also bring challenges to even experienced human interviewers in controlling the rhythm, that is, maintaining a balance between two conflicting aspects: (1) following the pre-constructed interview guide; (2) exploring the details that arise in the actual process to obtain additional information. This requirement requires fine control over the behavior of the interview agent. Following the well-known state, action, and reward reinforcement learning paradigm, the dialogue in the current interview is constructed as multiple rounds of dialogue in the current context content (i.e., context), action, and information process. Among them, the current context content refers to the natural language dialogue history alternating between the interviewee (interviewee) and the language model (i.e., the large language model called in the interview system). i When turning, use Qirepresents the questions asked by the interview system in this round (i.e., the interview interaction questions generated in each round of dialogue), and is represented by Ai Respondents indicated that Qi The answer (i.e. the historical interaction content input by the interviewee), the current context content (i.e. ) can be expressed as:

[0047] In the embodiments of the present invention, an action (i.e., a target action strategy) refers to the behavior of the agent (i.e., the language model called by the interview system) in the next question (i.e., the interview interaction question of the next round of dialogue) given the current context. By defining the action type and the conditions under which each action applies, the behavior of the language model can be fine-tuned to maximize the expected information collected. Specifically, each question of the language model Q (i.e., interview interaction questions) can be broken down into multiple actions in sequence i , j (i.e. at least one target action strategy):

[0048] Through the current context Select a series of actions i , j The process is a mapping, called the preset action mapping strategy ( P ):

[0049] In the embodiment of the present invention, according to the above definition, adjusting the behavior of the agent (i.e., the interviewee) involves defining an action space and establishing a strategy (i.e., an action mapping strategy). For the action space, in order to simultaneously satisfy the two conflicting aspects of the semi-structured interview, two actions are defined for each aspect. Figure 3 The strategy is encoded in the form of principles in corresponding prompt words (such as action mapping strategies), which specify behaviors and applicable conditions through a set of natural language guidelines.

[0050] like Figure 4 As shown in the figure, in the specific process of dialogue module processing: (1) it is necessary to pre-select and summarize classic literature and existing interview data, and obtain the strategy P used by the language model through existing prompt engineering technology. (2) In each round of dialogue in the current interview interaction process, the language model is called, and the interview guide information and current context content of the interview guide are input into the language model. (3) enabling the language model to process the acquired interview guide information and current context content, and selecting at least one specific action based on the interview guide information and current context content. i , j (i.e. at least one target action strategy), further i , j Combine into questions Qi (i.e., the next round of interview interaction questions). (4) Calling existing text-to-speech technology, such as Figure 4 The data conversion engine in the prior art generates questions Qi The voice is transmitted to the client. (5) Client display Qi , and play the voice. (6) Obtain the interviewee's input (i.e., the historical interaction content input by the interviewee) through client interaction: (a) Call the existing automatic speech recognition technology (Automatic Speech Recognition) to recognize and convert the interviewee's voice input into text; (b) Concatenate the above converted voice input with the user's text input as the interviewee's Qi Answer Ai (i.e., the historical interaction content input by the interviewee). (6) Repeat the question-and-answer process of (2) to (4) above until the language model determines that the interview has been completed (i.e., determines that the current interview interaction process is completed).

[0051] Step 103: Call the language model to analyze all the dialogue interaction contents to obtain dialogue interaction content analysis results.

[0052] In an embodiment of the present invention, the original output of the dialogue module in step 102 includes all the dialogue interaction content (i.e., a series of questions and answers). If no analysis is performed, these outputs cannot be effectively used in various interview application scenarios. The present invention can realize the analysis of all the dialogue interaction content from two dimensions, qualitative and quantitative. In the qualitative dimension process, the present invention can automatically summarize the summary of the dialogue information. Similar to the analysis memo, the summary contains the key points and main findings about the interview. In the quantitative dimension process, as the data scale increases, qualitative analysis becomes difficult, which usually leads to the loss of generality of the research. Therefore, a large language model is used to analyze the interview and obtain the score of the indicators proposed in the final stage of the guideline construction. Among them, the explanation of the score will be generated together with the score to enhance credibility. The present invention implements a specific process: the complete dialogue history (i.e., all the dialogue interaction content) obtained in the input dialogue module is embedded into a preset template to generate corresponding qualitative analysis prompt words and quantitative analysis prompt words. In the process of calling and guiding the generation of the large language model: the qualitative analysis results include the summary and main conclusions of the dialogue; the quantitative analysis results are analyzed according to each quantitative analysis indicator proposed in the guideline construction module, and the scores and reasons for the scores are given.

[0053] In addition, in an embodiment of the present invention, after calling the language model to analyze all the conversation interaction content and obtaining the conversation interaction content analysis results, the conversation interaction content and its corresponding conversation interaction content analysis results can be stored in a database, so that the conversation interaction content can be subsequently read from the database according to the conversation interaction content analysis results or the conversation interaction content analysis results can be obtained from the database and sent to the client for display in the form of a chart.

[0054] The present invention provides a method for processing interview interaction data based on knowledge guidance, which obtains the current context content in each round of dialogue in the current interview interaction process, performs splicing processing based on the current context content and the action mapping strategy, and obtains a first prompt word; the first prompt word is input into a language model to obtain at least one target action strategy adopted in the next round of dialogue output; the obtained interview guide information and at least one target action strategy are processed based on the language model to obtain the interview interaction questions of the next round of dialogue output by the language model; when it is determined based on the language model that the current interview interaction process is completed, all the dialogue interaction contents corresponding to the current interview interaction process are obtained, and all the dialogue interaction contents are analyzed to obtain the dialogue interaction content analysis results; the method can effectively improve the efficiency of interview interaction data processing, thereby greatly reducing the manpower cost.

[0055] The following is a description of the interview interaction data processing system based on knowledge guidance provided by the present invention. The interview interaction data processing system based on knowledge guidance described below and the interview interaction data processing method based on knowledge guidance described above can be referred to each other. Figure 6 As shown, it is a schematic diagram of the structure of the interview interaction data processing system based on knowledge guidance provided by the present invention. The interview interaction data processing system based on knowledge guidance provided by the present invention specifically includes the following parts: The prompt word acquisition module 601 is used to obtain the current context content in each round of dialogue in the current interview interaction process, and perform splicing processing based on the current context content and a preset action mapping strategy to obtain a first prompt word.

[0056] The interview interaction question acquisition module 602 is used to call a preset language model, input the first prompt word into the language model, and obtain at least one target action strategy to be adopted in the next round of dialogue output by the language model; based on the language model, the acquired interview guide information and the at least one target action strategy are processed to obtain the interview interaction question for the next round of dialogue output by the language model.

[0057] The dialogue interaction content acquisition module 603 is used to obtain all the dialogue interaction contents corresponding to the current interview interaction process output by the language model when it is determined that the current interview interaction process is completed based on the language model; wherein, all the dialogue interaction contents include interview interaction questions generated by at least one round of dialogue and historical interaction contents input by the interviewees.

[0058] The analysis module 604 is used to call the language model to analyze the entire dialogue interaction content to obtain a dialogue interaction content analysis result.

[0059] According to the knowledge-guided interview interaction data processing system of the present invention, before obtaining the current context content in each round of dialogue in the current interview interaction process, the system further includes: Guide building blocks for: Obtaining a target question input by an interviewee, performing splicing processing based on the target question and a preset question mapping strategy to obtain a second prompt word; calling a preset language model, inputting the second prompt word into the language model, and obtaining at least one interrogation question associated with the target question output by the language model; wherein the question mapping strategy includes a plurality of question mapping relationships corresponding to each target question; and the interrogation question is a branch question or an extended question based on the target question; Historical conversation data including the target question and the at least one probing question is obtained, corresponding interview guide information is obtained based on the historical conversation data including the target question and the at least one probing question, and the interview guide information is stored in a database.

[0060] According to the knowledge-guided interview interaction data processing system of the present invention, the target question and the preset question mapping strategy are spliced ​​together to obtain a second prompt word; a preset language model is called, the second prompt word is input into the language model, and at least one inquiry question associated with the target question output by the language model is obtained, specifically including: Performing a concatenation process based on the target question and the first question mapping relationship in the question mapping strategy to obtain a third prompt word; calling a preset language model, inputting the third prompt word into the language model, and obtaining a first analysis and processing result; Perform concatenation processing based on the target question and the second question mapping relationship in the question mapping strategy to obtain a fourth prompt word; call a preset language model, input the fourth prompt word into the language model, and obtain a second analysis processing result; Based on the target question and the third question mapping relationship in the question mapping strategy, a splicing process is performed to obtain a fifth prompt word; a preset language model is called, and the fifth prompt word is input into the language model to obtain at least one inquiry question associated with the target question; wherein the second prompt word includes the third prompt word, the fourth prompt word and the fifth prompt word; and the question mapping relationship includes a first question mapping relationship, a second question mapping relationship and a third question mapping relationship.

[0061] According to the knowledge-guided interview interaction data processing system of the present invention, before processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction questions for the next round of dialogue output by the language model, it also includes: an interview guide acquisition module, which is used to acquire the interview guide information corresponding to the current interview interaction process from the database.

[0062] According to the knowledge-guided interview interaction data processing system of the present invention, the interview interaction question obtaining module is specifically used for: The language model is called to combine the at least one target action strategy, and corresponding inquiry questions are generated based on the interview guide information to obtain interview interaction questions for the next round of dialogue output by the language model; wherein the at least one target action strategy includes at least one of a questioning action strategy, a responding action strategy, an inquiry action strategy, and a promotion action strategy.

[0063] According to the knowledge-guided interview interaction data processing system of the present invention, the dialogue interaction content acquisition module is specifically used for: Detecting an actual conversation turn of the current interview interaction process, and determining that the current interview interaction process has been completed when the actual conversation turn reaches a preset conversation turn threshold; After determining that the current interview interaction process has been completed, the dialogue interaction content of the current interview interaction process is summarized based on the language model to obtain all the dialogue interaction content corresponding to the current interview interaction process output by the language model.

[0064] In the embodiment of the present invention, Figure 5 The dialogue module shown includes the prompt word acquisition module 601, the interview interaction question acquisition module 602 and the dialogue interaction content acquisition module 603. Figure 5 In the interview system (i.e., the knowledge-guided interview interaction data processing system), an interview is created, the main questions are uploaded, the interviewee logs in to the interview system to participate in the current interview interaction, and the interview system analyzes the interview data and provides it to the researcher. In the internal service, the database is implemented using python's sqlite3. In the external service, the LLM API is a large model interface, and this interview system uses the GPT-4 large model; TTS (Text-To-Speech) is a text-to-speech module; ASR (Automatic Speech Recognition) is a speech-to-text module. The interview system is LM-Interview. The present invention is developed by using a knowledge-guided language model, and LM-Interview solves the problem through three modules (i.e., a guide construction module, a dialogue module, and an analysis module). Qualitative users can use the system to build an interview guide before the interview, and then collect a large amount of data through LLM-driven interviews without manual labor. Finally, insights are obtained from the analysis of the interview data by the interview system to improve the analysis efficiency. A method for automatically conducting semi-structured interviews using a knowledge-guided language model is proposed. Implemented LM-Interview, which aims to support qualitative users in the complete process of designing, conducting and analyzing semi-structured interviews.

[0065] like Figure 4 The figure shows how the various hardware architectures work together. During the guide construction module processing: (a) the user visits the page of the guide construction module; (b) the front end (i.e., the client) sends a GET request to the / create address to obtain the front end page and style; (c) fill in the main questions and send a POST request to the / create address; (d) the back end (i.e., the server) calls the language model to generate an interview guide; (e) inserts the data (i.e., the interview guide or interview guide information) into the database.

[0066] During the dialogue module processing: (f) The interviewee (i.e., the interviewee) visits the interview page, the front-end sends a GET request to the / interview address to obtain the front-end page and style, and the back-end creates an entry in the database to record the dialogue data (i.e., all dialogue interaction content). (g) The interviewee enters the answer (i.e., the interaction content entered by the interviewee) and sends a POST request to the / interview address; (h) The interview system asks a question: (1) The back-end searches for the corresponding dialogue data in the database; (2) The back-end calls the language model to generate the next question (i.e., the interview interaction question for the next round of dialogue); (3) The back-end calls the speech-to-text service to generate the voice of the question; (4) The back-end writes to the database and updates the dialogue data; (5) The back-end sends the generated question and voice to the front-end for display; (6) Repeat (g) and (h) until the interview dialogue interaction is completed.

[0067] During analysis module processing: (1) After the interviewee completes the interview, the analysis module automatically executes, calls the language model, analyzes the conversation data (i.e., all conversation interaction content), and stores the analysis (i.e., the results of the conversation interaction content analysis) in the database; (2) The researcher (i.e., the user) accesses the analysis page, and the front-end sends a GET request to the / analyze address to obtain the page content and style; (3) The back-end queries the database, reads the stored analysis data, and returns it to the front-end for display.

[0068] The knowledge-guided interview interaction data processing system provided by the present invention obtains the current context content in each round of dialogue in the current interview interaction process, performs splicing processing based on the current context content and the action mapping strategy, and obtains the first prompt word; inputs the first prompt word into the language model, and obtains at least one target action strategy adopted in the next round of dialogue output; processes the obtained interview guide information and at least one target action strategy based on the language model, and obtains the interview interaction questions of the next round of dialogue output by the language model; when it is determined based on the language model that the current interview interaction process is completed, obtains all the dialogue interaction contents corresponding to the current interview interaction process, and analyzes all the dialogue interaction contents to obtain the dialogue interaction content analysis results; it can effectively improve the efficiency of interview interaction data processing, thereby greatly reducing labor costs.

[0069] Figure 7 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 7As shown, the electronic device may include: a processor (processor) 701, a communication interface (Communications Interface) 704, a memory (memory) 702 and a communication bus 703, wherein the processor 701, the communication interface 704, and the memory 702 communicate with each other through the communication bus 703. The processor 701 can call the logic instructions in the memory 702 to execute the interview interaction data processing method based on knowledge guidance, which includes: obtaining the current context content in each round of dialogue in the current interview interaction process, and performing splicing processing based on the current context content and a preset action mapping strategy to obtain a first prompt word; calling a preset language model, inputting the first prompt word into the language model, and obtaining at least one target action strategy used in the next round of dialogue output by the language model; processing the obtained interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction question of the next round of dialogue output by the language model; when it is determined based on the language model that the current interview interaction process is completed, obtaining all the dialogue interaction content corresponding to the current interview interaction process output by the language model; wherein the all dialogue interaction content includes at least one interview interaction question generated by a round of dialogue and the historical interaction content input by the interviewee; calling the language model to analyze the all dialogue interaction content to obtain the dialogue interaction content analysis result.

[0070] In addition, the logic instructions in the above-mentioned memory 702 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk.

[0071] On the other hand, the present application also provides a computer program product, which includes a computer program, which can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the knowledge-guided interview interaction data processing method provided by the above methods, which includes: obtaining current context content in each round of dialogue in the current interview interaction process, and performing splicing processing based on the current context content and a preset action mapping strategy to obtain a first prompt word; calling a preset language model, inputting the first prompt word into the language model, and obtaining at least one target action strategy used in the next round of dialogue output by the language model; processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction question of the next round of dialogue output by the language model; when it is determined based on the language model that the current interview interaction process is completed, obtaining all dialogue interaction content corresponding to the current interview interaction process output by the language model; wherein the all dialogue interaction content includes at least one interview interaction question generated by a round of dialogue and the historical interaction content input by the interviewee; calling the language model to analyze the all dialogue interaction content to obtain a dialogue interaction content analysis result.

[0072] On the other hand, the present application also provides a computer-readable storage medium, the computer-readable storage medium includes a stored program, wherein the program, when running, executes the knowledge-guided interview interaction data processing method provided by the above methods, the method comprising: obtaining the current context content in each round of dialogue in the current interview interaction process, performing splicing processing based on the current context content and a preset action mapping strategy to obtain a first prompt word; calling a preset language model, inputting the first prompt word into the language model, and obtaining at least one target action strategy to be adopted in the next round of dialogue output by the language model; processing the obtained interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction question of the next round of dialogue output by the language model; when it is determined based on the language model that the current interview interaction process is completed, obtaining all the dialogue interaction content corresponding to the current interview interaction process output by the language model; wherein the all dialogue interaction content includes at least one interview interaction question generated by a round of dialogue and the historical interaction content input by the interviewee; calling the language model to analyze the all dialogue interaction content to obtain the dialogue interaction content analysis result.

[0073] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0074] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for processing interview interaction data based on knowledge guidance, characterized in that: include: In each round of dialogue in the current interview interaction process, current context content is obtained, and splicing processing is performed based on the current context content and a preset action mapping strategy to obtain a first prompt word; Calling a preset language model, inputting the first prompt word into the language model, and obtaining at least one target action strategy to be adopted in the next round of dialogue output by the language model; Processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain interview interaction questions for the next round of dialogue output by the language model; When it is determined based on the language model that the current interview interaction process is completed, obtaining all the dialogue interaction contents output by the language model and corresponding to the current interview interaction process; The entire dialogue interaction content includes at least one round of interview interaction questions generated by the dialogue and the historical interaction content input by the interviewee; The language model is called to analyze all the dialogue interaction contents to obtain a dialogue interaction content analysis result.

2. The method for processing interview interaction data based on knowledge guidance according to claim 1 is characterized in that: Before obtaining the current context content in each round of dialogue in the current interview interaction process, it also includes: Obtaining a target question input by an interviewee, performing splicing processing based on the target question and a preset question mapping strategy to obtain a second prompt word; calling a preset language model, inputting the second prompt word into the language model, and obtaining at least one interrogation question associated with the target question output by the language model; wherein the question mapping strategy includes a plurality of question mapping relationships corresponding to each target question; and the interrogation question is a branch question or an extended question based on the target question; Historical conversation data including the target question and the at least one probing question is obtained, corresponding interview guide information is obtained based on the historical conversation data including the target question and the at least one probing question, and the interview guide information is stored in a database.

3. The method for processing interview interaction data based on knowledge guidance according to claim 2 is characterized in that: The target question and the preset question mapping strategy are combined to obtain a second prompt word; Calling a preset language model, inputting the second prompt word into the language model, and obtaining at least one inquiry question associated with the target question output by the language model, specifically includes: Performing a concatenation process based on the target question and the first question mapping relationship in the question mapping strategy to obtain a third prompt word; calling a preset language model, inputting the third prompt word into the language model, and obtaining a first analysis and processing result; Perform concatenation processing based on the target question and the second question mapping relationship in the question mapping strategy to obtain a fourth prompt word; call a preset language model, input the fourth prompt word into the language model, and obtain a second analysis processing result; Based on the target question and the third question mapping relationship in the question mapping strategy, a splicing process is performed to obtain a fifth prompt word; a preset language model is called, and the fifth prompt word is input into the language model to obtain at least one inquiry question associated with the target question; wherein the second prompt word includes the third prompt word, the fourth prompt word and the fifth prompt word; and the question mapping relationship includes a first question mapping relationship, a second question mapping relationship and a third question mapping relationship.

4. The method for processing interview interaction data based on knowledge guidance according to claim 2 is characterized in that: Before processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction questions for the next round of dialogue output by the language model, it also includes: acquiring the interview guide information corresponding to the current interview interaction process from the database.

5. The method for processing interview interaction data based on knowledge guidance according to claim 1 is characterized in that: The step of processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain the interview interaction questions for the next round of dialogue output by the language model includes: The language model is called to combine the at least one target action strategy, and corresponding inquiry questions are generated based on the interview guide information to obtain interview interaction questions for the next round of dialogue output by the language model; wherein the at least one target action strategy includes at least one of a questioning action strategy, a responding action strategy, an inquiry action strategy, and a promotion action strategy.

6. The method for processing interview interaction data based on knowledge guidance according to claim 1 is characterized in that: In the case where it is determined based on the language model that the current interview interaction process is completed, obtaining all the dialogue interaction contents output by the language model and corresponding to the current interview interaction process specifically includes: Detecting an actual conversation turn of the current interview interaction process, and determining that the current interview interaction process has been completed when the actual conversation turn reaches a preset conversation turn threshold; After determining that the current interview interaction process has been completed, the dialogue interaction content of the current interview interaction process is summarized based on the language model to obtain all the dialogue interaction content corresponding to the current interview interaction process output by the language model.

7. A knowledge-guided interview interaction data processing system, characterized in that: include: A prompt word acquisition module is used to obtain the current context content in each round of dialogue in the current interview interaction process, and perform splicing processing based on the current context content and a preset action mapping strategy to obtain a first prompt word; An interview interaction question acquisition module, used to call a preset language model, input the first prompt word into the language model, and obtain at least one target action strategy to be adopted in the next round of dialogue output by the language model; Processing the acquired interview guide information and the at least one target action strategy based on the language model to obtain interview interaction questions for the next round of dialogue output by the language model; A dialogue interaction content acquisition module is used to obtain all dialogue interaction contents corresponding to the current interview interaction process output by the language model when it is determined that the current interview interaction process is completed based on the language model; wherein the all dialogue interaction contents include interview interaction questions generated by at least one round of dialogue and historical interaction contents input by the interviewee; The analysis module is used to call the language model to analyze all the dialogue interaction contents to obtain the dialogue interaction content analysis results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the knowledge-guided interview interaction data processing method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for processing interview interaction data based on knowledge guidance is implemented as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for processing interview interaction data based on knowledge guidance is implemented as described in any one of claims 1 to 6.

Images