Interview content regulation and control method and device based on multi-strategy cooperation

Abstract

The invention relates to the technical field of big data processing, and discloses an interview content regulation and control method and device based on multi-strategy collaboration.By implementing the method and device, the psychological state of an interviewee is evaluated through multi-dimensional parameters, the real-time interview state of the interviewee can be accurately grasped, and an accurate data basis is provided for subsequent personalized interview regulation and control; the interview content is intelligently analyzed from the macroscopic level and the microscopic level by adopting a double-strategy analysis model, so that the interview process is comprehensively optimized, and the interview quality and efficiency can be improved; meanwhile, based on the strategy analysis results output by the two strategy analysis models, the interview content can be flexibly changed according to the real-time state of the interviewee, the adaptability and flexibility of the interview are enhanced, and the success rate and effect of the interview are greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data processing, and particularly relates to a method and device for regulating interview content based on multi-strategy cooperation. BACKGROUND

[0002] With the development of big data models, their application approaches are becoming more and more extensive. For example, in the interview scene, big data models are also used to analyze interview data and give analysis suggestions, thereby assisting interviewers to adjust the interview. However, the existing big data models have many limitations in the application of interview adjustment. For example, in the evaluation of the psychological state of the interviewee, it often depends on a single factor, such as judging the participation degree of the interviewee only according to the answer content of the interviewee. The evaluation result obtained by the data analysis based on a single factor is necessarily not comprehensive and accurate, and it is difficult to accurately grasp the real conversation state of the interviewee in the interview process, for example, it is difficult to distinguish whether the topic participation degree of the interviewee decreases due to fatigue or lack of interest in the topic.

[0003] In addition, the traditional interview adopts a fixed mode process, and when facing the situation that the psychological state and reaction of the interviewee are constantly changing, it is difficult to obtain data in real time and quickly analyze and adjust the interview content, and it lacks flexibility and adaptability, and it is difficult to efficiently give reasonable and useful interview adjustment suggestions. SUMMARY

[0004] The present application provides a method and device for regulating interview content based on multi-strategy cooperation, which can evaluate the psychological state of the interviewee in real time and give scientific, reliable and practical interview adjustment suggestions.

[0005] The first aspect of the present application discloses a method for regulating interview content based on multi-strategy cooperation, the method comprising: obtaining target interview data corresponding to target interview content, the target interview data at least including a psychological state of an interviewee, first interview data for adjusting an interview direction and an interview stage, and second interview data for adjusting an interview action; the psychological state is obtained by comprehensively evaluating four interview parameters of the interviewee, i.e., trust degree, fatigue degree, participation degree and interest degree, which are calculated in advance; performing a first strategy analysis operation on the psychological state and the first interview data according to a pre-trained first strategy analysis model, to obtain a first strategy analysis result for the psychological state and the first interview data; the first strategy analysis result includes a global control action, and the global control action includes at least one action for performing interview stage conversion, interview rhythm adjustment and topic range change; performing a second strategy analysis operation on the psychological state and the second interview data according to the pre-trained second strategy analysis model, to obtain a second strategy analysis result for the psychological state and the second interview data; the second strategy analysis result comprises a target interview action, and the target interview action comprises an adjustment instruction for at least one of a questioning manner, a response manner, speech speed adjustment, and tone adjustment; performing an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result, to obtain an interview adjustment result for the target interview content.

[0006] As an optional implementation form, in the first aspect, the first strategy analysis model is a network model based on deep Q learning; the first interview data comprises first sub-data corresponding to a dialogue progress and a cumulative feature of dialogue history; the dialogue progress comprises a shallow interview stage or a deep interview stage, and further comprises a warm-up stage or a non-warm-up stage; the cumulative feature of dialogue history comprises at least one of a historical interaction mode, a historical interview rhythm, and a historical topic type; performing a first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model, to obtain a first strategy analysis result for the psychological state and the first interview data, comprises: constructing a first state vector corresponding to the psychological state and the first interview data according to the pre-trained first strategy analysis model; inputting the first state vector into the first strategy analysis model, and performing a first Q value prediction operation on the first state vector according to the first strategy analysis model, to obtain a plurality of first predicted Q values for the first state vector; each first predicted Q value corresponds to a first action, and each first action is used to adjust an interview direction, an interview stage, or an interview range of the target interview content; selecting a first target Q value adapted to the target interview content from all the first predicted Q values according to the first strategy analysis model and in combination with a pre-set first action selection strategy; determining the first target Q value and the first action corresponding thereto as the first strategy analysis result for the psychological state and the first interview data.

[0007] As an optional implementation form, in the first aspect, the first state vector comprises a first sub-vector corresponding to the dialogue progress and a second sub-vector corresponding to the cumulative feature of dialogue history. The method further comprises: performing a feature extraction and embedding operation on the first sub-vector according to the first strategy analysis model before inputting the first state vector into the first strategy analysis model, to obtain a feature extraction and embedding result for the first sub-vector; updating the first state vector according to the feature extraction and embedding result; the feature extraction and embedding operation is used to convert the first sub-vector from a discrete vector to a continuous vector.

[0008] As an optional implementation, in the first aspect of the present application, the second strategy analysis model is a deep Q-learning based network model; the second interview data includes second sub-data corresponding to last round dialogue feedback information, voice and emotion analysis result; the last round dialogue feedback information is used to at least indicate a feedback attitude of the interviewee for the last round dialogue, and the feedback attitude is a positive attitude or a negative attitude; the voice and emotion analysis result includes at least one feature of emotion, speech rate and pause corresponding to the voice of the interviewee; the second strategy analysis operation is performed on the psychological state and the second interview data according to the pre-trained second strategy analysis model, to obtain a second strategy analysis result for the psychological state and the second interview data, including: a second state vector corresponding to the psychological state and the second interview data is constructed according to the pre-trained second strategy analysis model; the second state vector is input into the second strategy analysis model, and a second Q value prediction operation is performed on the second state vector according to the second strategy analysis model, to obtain a plurality of second predicted Q values for the second state vector; each second predicted Q value corresponds to a second action, and each second action is used to adjust the interview action of the target interview content; a second target Q value adapted to the target interview content is selected from all the second predicted Q values according to the second strategy analysis model in combination with a pre-set second action selection strategy; the second target Q value and the second action corresponding thereto are determined as the second strategy analysis result for the psychological state and the second interview data.

[0009] As an optional implementation, in the first aspect of the present application, the interview action includes a first sub-action corresponding to a questioning manner and a response manner, and a second sub-action corresponding to a speech rate and a tone; each first sub-action belongs to a discrete and limited action category; the second sub-action belongs to a continuous action category or a discretizable action category; The questioning mode comprises any one of direct questioning, guided questioning and open questioning; the responding mode comprises any one of empathetic responding, feedback responding and analytical responding; all the second prediction Q values comprise first sub-prediction Q values corresponding to each first sub-action and second sub-prediction Q values corresponding to each second sub-action. The second sub-prediction Q value corresponding to each second sub-action is a single prediction continuous value for the speech speed / pitch, or a plurality of discrete prediction values.

[0010] As an optional implementation, in the first aspect of the present application, the method further comprises: According to the interview adjustment result, determine change information corresponding to the psychological state, the change information at least includes new parameter values corresponding to the trust degree, fatigue degree, participation degree and interest degree after the interview adjustment is performed; According to the psychological state and the change information, calculate parameter change information corresponding to each interview parameter, the parameter change information corresponding to each interview parameter comprises a parameter difference value corresponding to the interview parameter and a difference value type corresponding to the parameter difference value; the difference value type is used to indicate that the parameter difference value is a positive number or a negative number; Perform weighted summation operation on the parameter change information corresponding to each interview parameter and the weighting coefficient set for each interview parameter to obtain a first weighted summation result of the parameter change information corresponding to all the interview parameters; the first weighted summation result is used to perform model parameter optimization on the first strategy analysis model.

[0011] As an optional implementation, in the first aspect of the present application, the method further comprises: Obtain new interview content fed back by the interviewee after the interview adjustment operation is performed; Analyze the new interview content to obtain reward parameter information matched with a plurality of preset reward parameters, the reward parameter information comprises an emotional index corresponding to an emotional response parameter, a pause duration corresponding to a pause duration parameter, and a participation degree corresponding to an answer length and an answer depth; wherein the emotional index is used to indicate positive emotion or negative emotion of the interviewee; the pause duration is used to indicate a feedback duration of the interviewee triggering the new interview content; the participation degree is used to indicate content integrity and content complexity of the new interview content; Perform weighted summation operation on each reward parameter information and a weighting coefficient corresponding to each reward parameter information to obtain a second weighted summation result of all the reward parameter information; the second weighted summation result is used to perform model parameter optimization on the first strategy analysis model and the second strategy analysis model, respectively.

[0012] The second aspect of the present application discloses an interview content regulation device based on multi-strategy cooperation, which comprises: A data acquisition module is configured to acquire target interview data corresponding to target interview content, wherein the target interview data at least includes a psychological state of an interviewee, first interview data for adjusting an interview direction and an interview stage, and second interview data for adjusting an interview action; the psychological state is obtained by comprehensively evaluating four interview parameters of the interviewee, i.e., trust degree, fatigue degree, participation degree and interest degree, which are calculated in advance; A first strategy analysis module is configured to perform a first strategy analysis operation on the psychological state and the first interview data according to a pre-trained first strategy analysis model, so as to obtain a first strategy analysis result for the psychological state and the first interview data; the first strategy analysis result includes a global control action, and the global control action includes at least one action for performing an interview stage conversion, an interview rhythm adjustment and a topic range change; A second strategy analysis module is configured to perform a second strategy analysis operation on the psychological state and the second interview data according to a pre-trained second strategy analysis model, so as to obtain a second strategy analysis result for the psychological state and the second interview data; the second strategy analysis result includes a target interview action, and the target interview action includes an adjustment instruction for at least one action of a questioning manner, a response manner, a speech speed adjustment and a tone adjustment; An interview adjustment module is configured to perform an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result, so as to obtain an interview adjustment result for the target interview content.

[0013] As an optional implementation manner, in the second aspect of the present application, the first strategy analysis model is a network model based on deep Q learning; the first interview data includes first sub-data corresponding to cumulative features of a dialogue progress and a dialogue history; the dialogue progress includes a shallow interview stage or a deep interview stage, and further includes a warm-up stage or a non-warm-up stage; the history cumulative features include at least one feature of a history interaction mode, a history interview rhythm and a history topic type; The first strategy analysis module performs a first strategy analysis operation on the psychological state and the first interview data according to a pre-trained first strategy analysis model, so as to obtain a first strategy analysis result for the psychological state and the first interview data, and the manner specifically includes: According to the pre-trained first strategy analysis model, a first state vector corresponding to the psychological state and the first interview data is constructed; input the first state vector into the first strategy analysis model, and perform a first Q value prediction operation on the first state vector according to the first strategy analysis model to obtain a plurality of first predicted Q values for the first state vector; each of the first predicted Q values corresponds to a first action, and each of the first actions is used to adjust an interview direction, an interview stage or an interview range of the target interview content; According to the first strategy analysis model, a first target Q value that is adapted to the target interview content is selected from all the first predicted Q values in combination with a preset first action selection strategy. The first target Q value and the corresponding first action are determined as the first strategy analysis result for the psychological state and the first interview data.

[0014] As an optional implementation, in the second aspect of the present application, the first state vector includes a first sub-vector corresponding to the conversation progress and a second sub-vector corresponding to accumulated features of the conversation history. The manner in which the first strategy analysis module performs the first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model to obtain the first strategy analysis result for the psychological state and the first interview data specifically includes: Before inputting the first state vector into the first strategy analysis model, a feature extraction and embedding operation is performed on the first sub-vector according to the first strategy analysis model to obtain a feature extraction and embedding result for the first sub-vector. The first state vector is updated according to the feature extraction and embedding result; the feature extraction and embedding operation is used to convert the first sub-vector from a discrete vector to a continuous vector.

[0015] As an optional implementation, in the second aspect of the present application, the second strategy analysis model is a network model based on deep Q learning; the second interview data includes a second sub-data corresponding to last round conversation feedback information, voice and emotion analysis results; the last round conversation feedback information is used to at least indicate a feedback attitude of the interviewee for the last round conversation, and the feedback attitude is a positive attitude or a negative attitude; the voice and emotion analysis results include at least one feature of emotion, speech speed and pause corresponding to the voice of the interviewee; The manner in which the first strategy analysis module performs the second strategy analysis operation on the psychological state and the second interview data according to the pre-trained second strategy analysis model to obtain the second strategy analysis result for the psychological state and the second interview data specifically includes: According to the pre-trained second strategy analysis model, a second state vector corresponding to the psychological state and the second interview data is constructed. inputting the second state vector into the second strategy analysis model, and performing a second Q value prediction operation on the second state vector according to the second strategy analysis model to obtain a plurality of second predicted Q values for the second state vector; each of the second predicted Q values corresponds to a second action, and each of the second actions is used to adjust an interview action of the target interview content; selecting, according to the second strategy analysis model and in combination with a preset second action selection strategy, a second target Q value that is adapted to the target interview content from all the second predicted Q values; determining the second target Q value and the second action corresponding thereto as a second strategy analysis result for the psychological state and the second interview data.

[0016] As an optional implementation form, in the second aspect of the present application, the interview action includes a first sub-action corresponding to a question-asking manner and a response manner, and a second sub-action corresponding to a speech speed and a speech tone; each of the first sub-actions belongs to a discrete and limited action category; and the second sub-action belongs to a continuous action category or a discretizable action category. The question-asking manner includes any one of a direct question-asking, a guided question-asking and an open question-asking; the response manner includes any one of an empathetic response, a feedback response and an analytical response; and all the second predicted Q values include a first sub-predicted Q value corresponding to each of the first sub-actions and a second sub-predicted Q value corresponding to each of the second sub-actions. The second sub-predicted Q value corresponding to each of the second sub-actions is a single predicted continuous value for the speech speed / speech tone, or a plurality of discrete predicted values.

[0017] As an optional implementation form, in the second aspect of the present application, the device further includes: a determination module configured to determine, according to the interview adjustment result, change information corresponding to the psychological state, the change information at least including new parameter values corresponding to the four interview parameters of the trust degree, the fatigue degree, the participation degree and the interest degree after the interview adjustment is performed; a calculation module configured to calculate, according to the psychological state and the change information, parameter change information corresponding to each of the interview parameters, the parameter change information corresponding to each of the interview parameters including a parameter difference value corresponding to the interview parameter and a difference value type corresponding thereto; and the difference value type is used to indicate whether the parameter difference value is positive or negative. The first weighted sum module is configured to perform a weighted sum operation on the parameter change information corresponding to each of the interview parameters and a weighting coefficient set for each of the interview parameters, to obtain a first weighted sum result of the parameter change information corresponding to all the interview parameters; and the first weighted sum result is used for model parameter optimization of the first strategy analysis model.

[0018] As an optional implementation, in the second aspect of the present application, the data acquisition module is further configured to acquire new interview content fed back by the interviewee after the interview adjustment operation is performed. The device further comprises: The analysis module is configured to analyze the new interview content, to obtain reward parameter information matched with a plurality of preset reward parameters, the reward parameter information comprising an emotional index corresponding to an emotional response parameter, a pause duration corresponding to a pause duration parameter, and an engagement corresponding to an answer length and an answer depth; wherein the emotional index is used to indicate whether the interviewee is in a positive emotion or a negative emotion; the pause duration is used to indicate a feedback duration of the interviewee when triggering the new interview content; and the engagement is used to indicate a content completeness and a content complexity of the new interview content. The second weighted sum module is configured to perform a weighted sum operation on each of the reward parameter information and a weighting coefficient set for each of the reward parameter information, to obtain a second weighted sum result of all the reward parameter information; and the second weighted sum result is used for model parameter optimization of the first strategy analysis model and the second strategy analysis model respectively.

[0019] The third aspect of the present application discloses another interview content regulation device based on multi-strategy cooperation, the device comprising: A memory storing executable program codes; A processor coupled with the memory; The processor invokes the executable program codes stored in the memory to execute part or all of the steps of the interview content regulation method based on multi-strategy cooperation according to any one of the first aspect of the present application.

[0020] The fourth aspect of the present application discloses a computer storage medium storing computer instructions, the computer instructions being invoked to execute part or all of the steps of the interview content regulation method based on multi-strategy cooperation according to any one of the first aspect of the present application.

[0021] Compared with the prior art, the present application has the following beneficial effects: In the embodiment of the present application, a multi-strategy cooperative interview content regulation method is provided, which comprises the following steps: obtaining target interview data corresponding to target interview content, wherein the target interview data at least comprises a psychological state of an interviewee, first interview data for adjusting an interview direction and an interview stage, and second interview data for adjusting an interview action; the psychological state is obtained by comprehensively evaluating four interview parameters of the interviewee, i.e., trust degree, fatigue degree, participation degree and interest degree, which are calculated in advance; performing a first strategy analysis operation on the psychological state and the first interview data according to a pre-trained first strategy analysis model to obtain a first strategy analysis result for the psychological state and the first interview data; the first strategy analysis result comprises a global control action, and the global control action comprises at least one action for performing an interview stage conversion, an interview rhythm adjustment and a topic range change; performing a second strategy analysis operation on the psychological state and the second interview data according to a pre-trained second strategy analysis model to obtain a second strategy analysis result for the psychological state and the second interview data; the second strategy analysis result comprises a target interview action, and the target interview action comprises an adjustment instruction for at least one action of a questioning manner, a response manner, a speech speed adjustment and a tone adjustment; and performing an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result to obtain an interview adjustment result for the target interview content. It can be seen that, by evaluating the psychological state of the interviewee from multiple dimensions, the real-time interview state of the interviewee can be accurately grasped, and accurate data basis is provided for subsequent personalized interview regulation; by using the double strategy analysis models, the interview process is comprehensively optimized from macro and micro aspects, which is beneficial to improving the interview quality and efficiency; meanwhile, based on the strategy analysis results output by the two strategy analysis models, the interview content can be flexibly changed according to the real-time state of the interviewee, the adaptability and flexibility of the interview are enhanced, and the success rate and effect of the interview are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort.

[0023] Figure 1 is a flowchart of a multi-strategy cooperative interview content regulation method disclosed by the embodiment of the present application; Figure 2 is a flowchart of another multi-strategy cooperative interview content regulation method disclosed by the embodiment of the present application; Figure 3is a structural schematic view of an interview content regulation device based on multi-strategy cooperation disclosed by an embodiment of the present application; Figure 4 is a structural schematic view of another interview content regulation device based on multi-strategy cooperation disclosed by an embodiment of the present application; Figure 5 is a structural schematic view of still another interview content regulation device based on multi-strategy cooperation disclosed by an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0025] The terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish different objects, and are not used to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or end including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product, or end.

[0026] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0027] The present application discloses an interview content regulation method and device based on multi-strategy cooperation. The psychological state of the interviewee is evaluated through multi-dimensional parameters, which can accurately grasp the real-time interview state of the interviewee, providing accurate data basis for subsequent personalized interview regulation. The interview process is comprehensively optimized from the macro and micro levels by using a double-strategy analysis model, which is beneficial to improving the interview quality and efficiency. At the same time, based on the strategy analysis results output by the two strategy analysis models, the interview content can be flexibly changed according to the real-time state of the interviewee, enhancing the adaptability and flexibility of the interview, greatly improving the success rate and effect of the interview. The following will be described in detail.

[0028] Embodiment one Please refer to Figure 1 , Figure 1 is a flowchart of a method for regulating interview content based on multi-strategy cooperation according to an embodiment of the present application. In the method, Figure 1 The method for regulating interview content based on multi-strategy cooperation can be applied to a device for regulating interview content based on multi-strategy cooperation, and the present application is not limited to this. As shown in Figure 1 The method for regulating interview content based on multi-strategy cooperation can include the following operations: 101. Obtain target interview data corresponding to target interview content, wherein the target interview data at least includes a psychological state of an interviewee, first interview data for adjusting an interview direction and an interview stage, and second interview data for adjusting an interview action.

[0029] In the present application, the psychological state is obtained by comprehensively evaluating the four interview parameters of the interviewee, i.e., trust degree, fatigue degree, participation degree, and interest degree.

[0030] In the present application, the psychological state can be directly obtained by using the data corresponding to the four interview parameters. In this case, each interview parameter corresponds to a numerical value, which can be a percentage value, used to indicate the “degree value” of the interview parameter. For example, a trust degree of 30% indicates that the current interviewee has a low degree of trust in the interviewer. For another example, a fatigue degree of 70% indicates that the current interviewee is tired of the current interview content, which can be caused by the long duration of the interview.

[0031] In the present application, the data corresponding to the four interview parameters can be optionally weighted and summed, and then the weighted sum result is compared with a plurality of predetermined psychological states (each psychological state corresponds to a score value, such as [0.3, 0.4] corresponding to an interview fatigue state of the interviewee, and [0.7, 0.8] corresponding to a high interest state of the interviewee), to obtain a target psychological state matching the weighted sum result.

[0032] In the present application, the psychological state is evaluated by using multi-dimensional parameters, which can more comprehensively and accurately reflect the real state of the interviewee during the interview. For example, the participation degree of the interviewee may be biased if it is determined only according to the answer content of the interviewee, but the participation degree of the interviewee can be more accurately determined by combining the trust degree, fatigue degree, and other parameters. Based on such accurate psychological state evaluation, a reliable basis can be provided for subsequent interview regulation, and individualized interview adjustment can be realized according to the individual characteristics of different interviewees.

[0033] 102. performing a first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model, to obtain a first strategy analysis result for the psychological state and the first interview data.

[0034] In the embodiment of the present application, the first strategy analysis result includes a global control action, and the global control action includes at least one action for performing interview stage conversion, interview rhythm adjustment, and topic range change.

[0035] In the embodiment of the present application, the first strategy analysis result obtained based on the first strategy analysis model can control the overall trend of the interview from a macro perspective, ensure that the interview proceeds in an orderly manner according to the predetermined target, and avoid the interview from deviating from the theme or falling into a stalemate. For example, when it is found that the interviewee has a reduced interest in the current topic, the topic range can be timely converted through the global control action to maintain the activity of the interview.

[0036] 103. performing a second strategy analysis operation on the psychological state and the second interview data according to the pre-trained second strategy analysis model, to obtain a second strategy analysis result for the psychological state and the second interview data.

[0037] In the embodiment of the present application, the second strategy analysis result includes a target interview action, and the target interview action includes an adjustment instruction for at least one of the following: a questioning manner, a response manner, a speed adjustment, and a tone adjustment.

[0038] In the embodiment of the present application, the second strategy analysis result obtained based on the second strategy analysis model can optimize the specific operation of the interview from a micro perspective and improve the interaction effect of the interview and the experience of the interviewee. For example, according to the psychological state of the interviewee, the questioning manner is adjusted, and a more gentle or more heuristic manner is used to ask questions, so that the interviewee is more willing to share information.

[0039] In the embodiment of the present application, through the cooperative work of the two strategy analysis models, comprehensive interview process optimization is performed from the macro and micro perspectives, which is beneficial to improve the quality and efficiency of the interview.

[0040] 104. performing an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result, to obtain an interview adjustment result for the target interview content.

[0041] In the embodiment of the present application, through steps 101-104, a real-time and dynamic adjustment mechanism for the target interview content is set, and based on the mechanism, changes can be made in a timely manner according to the real-time state and feedback of the interviewee.

[0042] In the actual interview process, the psychological state and reaction of the interviewee are constantly changing, and the traditional fixed mode of the interview process is difficult to adapt to such changes. In contrast, through the real-time and dynamic adjustment mechanism set by steps 101-104, relevant data can be obtained and quickly analyzed in real time, and the interview content can be quickly adjusted according to the analysis results, such as adjusting the interview stage, rhythm, topic range, and specific interview actions.

[0043] Specifically, in a certain interview scenario, if the fatigue of the interviewee is found to be rising during the interview process by real-time collection and analysis of the target interview data, and based on the first and second strategy analysis results given by the two strategy analysis models, it is concluded that the interview rhythm needs to be adjusted in a timely manner, then the interviewer can slow down the speaking speed, lower the tone, or appropriately shorten the interview time of the current stage according to the suggestion, thereby improving the flexibility and adaptability of the interview, and ensuring that the interview can be successfully completed and achieve the expected goal.

[0044] It can be seen that the implementation Figure 1 The interview content regulation method based on multi-strategy cooperation described above can accurately grasp the real-time interview state of the interviewee by evaluating the psychological state of the interviewee through multi-dimensional parameters, and provide accurate data basis for subsequent personalized interview regulation; the double-strategy analysis model is used to comprehensively optimize the interview process from the macro and micro levels, which is beneficial to improve the interview quality and efficiency; at the same time, based on the strategy analysis results output by the two strategy analysis models, the interview content can be flexibly changed according to the real-time state of the interviewee, which enhances the adaptability and flexibility of the interview, and greatly improves the success rate and effect of the interview.

[0045] In an optional embodiment, the first strategy analysis model is a network model based on deep Q learning; the first interview data includes first sub-data corresponding to the cumulative characteristics of the dialogue progress and the dialogue history; the dialogue progress includes a shallow interview stage or a deep interview stage, and also includes a warm-up stage or a non-warm-up stage; the historical cumulative characteristics include at least one of historical interaction mode, historical interview rhythm, and historical topic type.

[0046] In the optional embodiment, the deep Q learning combines the feature extraction capability of deep learning and the decision-making capability of reinforcement learning. In the interview regulation scene, a large amount of interview data and experience can be learned to automatically learn how to make the optimal interview direction, stage or range adjustment decision under different psychological states and first interview data combinations. For example, under different dialogue progress (such as a shallow interview stage or a deep interview stage, a warm-up stage or a non-warm-up stage) and different dialogue history cumulative features (such as a historical interaction mode, a historical interview rhythm, a historical topic type, etc.), the model can accurately determine the most suitable adjustment action according to the knowledge learned by pre-training, so that compared with the traditional rule-based or simple statistical method, the output decision of the first strategy analysis model using deep Q learning is more scientific and accurate.

[0047] In the optional embodiment, the dialogue progress can be used to reflect the current stage and needs of the interview, for example, in the warm-up stage, a lighter topic and interaction mode can be used. The cumulative features of the dialogue history can include various information in the past interview process, wherein the historical interaction mode can be used to reflect the communication habits between the interviewee and the interviewer; the historical interview rhythm can be used to reflect the preference of the interview progress speed; and the historical topic type can understand the topic field of interest of the interviewee.

[0048] In the optional embodiment, by fusing the two types of information to construct the first state vector, the current state of the interview can be more comprehensively and accurately described, and more rich and accurate data basis can be provided for subsequent decision-making.

[0049] In the optional embodiment, the manner of step 102 performing the first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model to obtain the first strategy analysis result for the psychological state and the first interview data specifically includes: constructing a first state vector corresponding to the psychological state and the first interview data according to the pre-trained first strategy analysis model; inputting the first state vector into the first strategy analysis model, and performing a first Q value prediction operation on the first state vector according to the first strategy analysis model to obtain a plurality of first predicted Q values for the first state vector; each first predicted Q value corresponds to a first action, and each first action is used to adjust the interview direction or the interview stage or the interview range of the target interview content; selecting a first target Q value adapted to the target interview content from all the first predicted Q values according to the first strategy analysis model in combination with a pre-set first action selection strategy; determining the first target Q value and the first action corresponding thereto as the first strategy analysis result for the psychological state and the first interview data.

[0050] In the optional embodiment, the first state vector includes a first sub-vector corresponding to the progress of the dialogue and a second sub-vector corresponding to the accumulated features of the dialogue history.

[0051] In the optional embodiment, the first action selection strategy can adopt an epsilon-greedy strategy.

[0052] In the optional embodiment, when the first Q value prediction operation is performed, the processing can be performed by a multi-layer perception (MLP) or other structure (such as a convolutional layer or an LSTM layer) included in the first strategy analysis model. For the model structure of the first strategy analysis model, refer to a conventional deep Q learning (DQN) network, which is not described in detail in the optional embodiment.

[0053] In the optional embodiment, the method further includes: Before the first state vector is input into the first strategy analysis model, performing a feature extraction and embedding operation on the first sub-vector according to the first strategy analysis model to obtain a feature extraction and embedding result for the first sub-vector; updating the first state vector according to the feature extraction and embedding result; the feature extraction and embedding operation is used to convert the first sub-vector from a discrete vector to a continuous vector.

[0054] In the optional embodiment, the first sub-data corresponding to the progress of the dialogue and the accumulated features of the dialogue history are usually a large number of discrete vectors, and the discrete vectors have certain limitations in expressing information and are difficult to fully capture the complex relationships and subtle differences between data. After being converted into continuous vectors, the features and semantic information of the data can be better represented, so that the first strategy analysis model can more effectively learn and process the data.

[0055] In the optional embodiment, a prediction and selection mechanism based on Q value is set for the first strategy analysis model, which can ensure that the selected action can maximize the optimization of the interview direction, stage or range in the current state, and improve the efficiency and effectiveness of the interview. For example, in a certain interview stage, the model predicts multiple possible adjustment actions and their corresponding Q values, and by selecting the action corresponding to the target Q value, blind adjustment can be avoided, and the interview can develop in a more favorable direction.

[0056] It can be seen that in the optional embodiment, when constructing the to-be-processed data (the first state vector), the progress of the dialogue and the historical cumulative features can be comprehensively considered, and the feature extraction and embedding conversion are performed on the related sub-vectors, the discrete vectors are converted into continuous vectors convenient for processing by the model, and the processing capability of the model is improved. Then, the first strategy analysis model based on deep Q learning is adopted, the Q value prediction and selection mechanism are adopted, and the first action output by the first strategy analysis model is ensured to adapt to the target interview content, so that the first strategy analysis result output by the first strategy analysis model can realize the accurate and scientific optimization of the interview direction, stage and range, and improve the efficiency and quality of the interview.

[0057] In another optional embodiment, the second strategy analysis model is a network model based on deep Q learning; the second interview data includes second sub-data corresponding to the feedback information of the last round of dialogue, voice and emotional analysis results; the feedback information of the last round of dialogue is used at least to indicate the feedback attitude of the interviewee to the last round of dialogue, and the feedback attitude is a positive attitude or a negative attitude; the voice and emotional analysis results include at least one of the features of emotion, speech rate and pause corresponding to the voice of the interviewee.

[0058] In the optional embodiment, the feedback attitude provided by the feedback information of the last round of dialogue can be used to let the interviewer understand the acceptance degree of the interviewee to the target interview content, so as to adjust the interview content appropriately. For example, if the feedback attitude of the interviewee is negative, the interviewer may need to adjust the questioning method or topic. The voice and emotional analysis results cover the features such as emotion, speech rate and pause corresponding to the voice of the interviewee, and the inner state of the interviewee can be analyzed in detail by analyzing these features. For example, the accelerated speech rate may indicate that the interviewee is excited or nervous; the increased pause may mean that the interviewee is thinking or hesitating. Through the integration of multi-dimensional data, the interview situation can be more comprehensively and accurately described, and rich and accurate decision data can be provided for the second strategy analysis model.

[0059] In the optional embodiment, the manner of performing the second strategy analysis operation on the psychological state and the second interview data according to the pre-trained second strategy analysis model to obtain the second strategy analysis result for the psychological state and the second interview data in the above-mentioned step 103 specifically includes: constructing a second state vector corresponding to the psychological state and the second interview data according to the pre-trained second strategy analysis model; inputting the second state vector into the second strategy analysis model, and performing a second Q value prediction operation on the second state vector according to the second strategy analysis model to obtain a plurality of second predicted Q values for the second state vector; each second predicted Q value corresponds to a second action, and each second action is used to adjust the interview action of the target interview content; According to the second strategy analysis model, a second target Q value adapted to the target interview content is selected from all the second predicted Q values in combination with a preset second action selection strategy; The second target Q value and the corresponding second action are determined as the second strategy analysis result for the psychological state and the second interview data.

[0060] In this optional embodiment, the interview action includes a first sub-action corresponding to the questioning manner and the response manner, and a second sub-action corresponding to the speech speed and the tone; each first sub-action belongs to a discrete and limited action category; and the second sub-action belongs to a continuous action category or a discretizable action category.

[0061] In this optional embodiment, the questioning manner includes any one of direct questioning, guided questioning and open questioning; the response manner includes any one of empathetic response, feedback response and analytical response; and all the second predicted Q values include a first sub-predicted Q value corresponding to each first sub-action and a second sub-predicted Q value corresponding to each second sub-action.

[0062] In this optional embodiment, the second action selection strategy can adopt an epsilon-greedy strategy.

[0063] In this optional embodiment, similar to the above description of the first strategy analysis model, when the second Q value prediction operation is performed, it can be processed by a multi-layer perception (MLP) or other structures (such as convolutional layer, LSTM layer) included in the second strategy analysis model, and the model structure of the second strategy analysis model can refer to a conventional deep Q learning (DQN) network, which is not described in detail in this optional embodiment.

[0064] In this optional embodiment, the second sub-predicted Q value corresponding to each second sub-action is a single predicted continuous value for the speech speed / tone, or a plurality of discrete predicted values.

[0065] In this optional embodiment, the second strategy analysis model also sets a prediction and selection mechanism based on the Q value, which can ensure that the selected second action can maximize the optimization of the interview action in the current state, and improve the interaction effect of the interview and the experience of the interviewee. For example, at a certain interview moment, the second strategy analysis model predicts a plurality of possible questioning manners, response manners, speech speed and tone adjustment actions and corresponding Q values, and by selecting the action corresponding to the target Q value, blind adjustment can be avoided, so that the interview action is more in line with the needs of the interviewee and the interview target.

[0066] It can be seen that in the optional embodiment, the required key information can be integrated when the second state vector is constructed, the data quality of the input second strategy analysis model is improved, and the analysis efficiency of the second strategy analysis model for the second state vector is also improved. Further, the output second action is ensured to adapt to the target interview content through Q value prediction and selection, that is, the determination of the second action is scientific and accurate. In addition, each second action is intelligently subdivided according to the action category, including discrete, first sub-actions convenient for quick selection, and also including continuous or discretizable, second sub-actions capable of accurately adjusting the speech speed and tone. The scientific and refined second action is taken as the second strategy analysis result for subsequent interview adjustment, which is beneficial to improve the interaction effect and the interviewee experience of the interview, and realizes more intelligent and efficient interview action regulation.

[0067] Embodiment two Please refer to Figure 2 , Figure 2 is another flowchart of a method for interview content regulation based on multi-strategy cooperation according to an embodiment of the present application. In the method, Figure 2 The method for interview content regulation based on multi-strategy cooperation described above can be applied to a device for interview content regulation based on multi-strategy cooperation, and the embodiments of the present application are not limited thereto. As shown in Figure 2 The method for interview content regulation based on multi-strategy cooperation can include the following operations: 201. Obtain target interview data corresponding to target interview content, wherein the target interview data at least includes the psychological state of an interviewee, first interview data for adjusting the direction and stage of an interview, and second interview data for adjusting an interview action.

[0068] 202. Perform a first strategy analysis operation on the psychological state and the first interview data according to a pre-trained first strategy analysis model, to obtain a first strategy analysis result for the psychological state and the first interview data.

[0069] 203. Perform a second strategy analysis operation on the psychological state and the second interview data according to a pre-trained second strategy analysis model, to obtain a second strategy analysis result for the psychological state and the second interview data.

[0070] 204. Perform an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result, to obtain an interview adjustment result for the target interview content.

[0071] In the embodiments of the present application, other descriptions of steps 201-204 can refer to other specific descriptions of steps 101-104 in Embodiment One, and the embodiments of the present application will not be repeated here.

[0072] 205、adjust the result according to the interview, determine the change information corresponding to the psychological state.

[0073] In the embodiment of the present application, the change information at least includes new parameter values corresponding to the four interview parameters of trust degree, fatigue degree, participation degree and interest degree after the interview adjustment is performed. The four interview parameters directly reflect the psychological state of the interviewee and the degree of investment in the interview.

[0074] 206、according to the psychological state and the change information, calculate the parameter change information corresponding to each interview parameter.

[0075] In the embodiment of the present application, the parameter change information corresponding to each interview parameter includes a parameter difference value corresponding to the interview parameter and a difference value type corresponding to the parameter difference value; the difference value type is used to indicate that the parameter difference value is a positive number or a negative number.

[0076] In the embodiment of the present application, the parameter difference value can directly show the change range of the parameter before and after the interview adjustment is performed; and the difference value type corresponding to the parameter difference value is used to clearly indicate the direction of the parameter change. The parameter change information corresponding to each interview parameter is obtained by calculating the difference between the new parameter value and the original parameter value. For example, for the trust degree, the new trust degree is subtracted from the original trust degree to obtain the trust degree difference value, and the trust degree difference value can be a positive number or a negative number. Further, if the parameter difference value of the trust degree is positive, it indicates that the trust degree of the interviewee to the interviewer is improved after the interview adjustment is performed, otherwise, if the parameter difference value of the trust degree is negative, it indicates that the trust degree of the interviewee to the interviewer is decreased after the interview adjustment is performed. By setting a clear calculation method for the parameter change information, the influence of the interview adjustment on the parameters of the psychological state of the interviewee can be more comprehensively and accurately reflected, and more targeted basis is provided for subsequent model optimization.

[0077] 207、perform weighted summation operation on the parameter change information corresponding to each interview parameter and the weighting coefficient set for each interview parameter to obtain a first weighted summation result of the parameter change information corresponding to all interview parameters.

[0078] In the embodiment of the present application, the first weighted summation result is used to perform model parameter optimization on the first strategy analysis model.

[0079] In the embodiment of the present application, different interview parameters may have different importance in the actual interview scene in terms of influencing the interview effect and the psychological state of the interviewee. By setting the weighting coefficients, different weights can be given to different parameters according to the actual situation. For example, in some interview scenes, trustworthiness may be more critical to the smooth progress of the interview and the acquisition of effective information, so a higher weighting coefficient can be set for it; while the influence of fatigue is relatively small, the weighting coefficient can be set lower. Or, when the conversation time is too long and the fatigue value accumulates too high, the weighting coefficient corresponding to the fatigue can be increased, and so on, which is not limited in the embodiment of the present application.

[0080] It can be seen that the implementation Figure 2 The interview content regulation method based on multi-strategy cooperation described herein accurately determines the new values of the four key parameters (trustworthiness, fatigue, participation, and interest) of the interview adjusted psychological state, and has scientific and quantitative calculation parameter change information (including difference and difference type), and then obtains a comprehensive evaluation result through weighted summation operation, which is used to optimize the first strategy analysis model parameters. Through this process, the interviewee's acceptance of the interview adjustment result can be scored in a timely manner, which is also used as auxiliary optimization data for the first strategy analysis model, that is, a timely detection mechanism for the interview adjustment result and an optimization mechanism for the first strategy analysis model are established, so that the first strategy analysis model optimized subsequently can better fit the psychological state of the interviewee, thereby effectively improving the pertinence and quality of the interview and achieving more intelligent interview content regulation.

[0081] In an optional embodiment, the method further comprises: obtaining new interview content fed back by the interviewee after performing the interview adjustment operation; analyzing the new interview content to obtain reward parameter information matched with the plurality of preset reward parameters, the reward parameter information including an emotional index corresponding to an emotional response parameter, a pause duration corresponding to a pause duration parameter, and a participation degree corresponding to an answer length and an answer depth; wherein the emotional index is used to indicate whether the interviewee is in a positive emotion or a negative emotion; the pause duration is used to indicate the feedback duration of the interviewee triggering the new interview content; and the participation degree is used to indicate the content completeness and content complexity of the new interview content; performing a weighted summation operation on each reward parameter information and the weighting coefficient corresponding to each reward parameter information to obtain a second weighted summation result for all reward parameter information; the second weighted summation result is used to respectively optimize the model parameters of the first strategy analysis model and the second strategy analysis model.

[0082] In the optional embodiment, the new interview content is the direct feedback given by the interviewee based on the adjusted interview strategy, which can truly reflect the interviewee's acceptance, understanding and willingness to participate in the current interview, and provide comprehensive and original data sources for subsequent analysis. For example, if the interviewee gives more detailed and in-depth answers after the interview is adjusted, it means that the adjusted strategy may be more in line with the needs of the interviewee and stimulate their desire to express.

[0083] In the optional embodiment, since the importance of different reward parameters in influencing the interview effect and model performance may be different, by setting reasonable weighting coefficients, different weights can be given to different parameters according to actual conditions. For example, in some interview scenarios, emotional response may be more critical to establishing a good interview relationship, so a higher weighting coefficient can be set for the emotional index; and in interviews that require a deep understanding of the interviewee's knowledge level and thinking ability, participation may be more important, and its weighting coefficient can be correspondingly increased.

[0084] In the optional embodiment, the calculation of the second weighted sum result can comprehensively consider the influence of each reward parameter, which is conducive to improving the comprehensiveness and accuracy of the evaluation of the interview adjustment result. Then, using this comprehensive and accurate second weighted sum result to optimize the model parameters of the first strategy analysis model and the second strategy analysis model respectively can make the two models better adjust the interview strategy according to the feedback of the interviewee in subsequent interviews, which is conducive to improving the pertinence and effectiveness of subsequent interviews, and realizing more intelligent and accurate interview content regulation.

[0085] As can be seen, in the optional embodiment, the new content of the interviewee's feedback after the interview is adjusted is first obtained, and then the emotional index, pause duration, participation and other quantitative reward parameter information are analyzed in detail, and a comprehensive evaluation result is obtained by weighting and summing with reasonable weighting coefficients. Using this result to optimize the parameters of the first and second strategy analysis models can comprehensively and accurately evaluate the interview adjustment effect, so that the first and second strategy analysis models are more in line with the feedback adjustment strategy of the interviewee, while improving the pertinence and effectiveness of the interview.

[0086] Embodiment three Please refer to Figure 3 , Figure 3 is a structural schematic diagram of an interview content regulation device based on multi-strategy cooperation disclosed by the embodiments of the present application. The interview content regulation device based on multi-strategy cooperation can be an interview content regulation terminal, device, system or server based on multi-strategy cooperation. The server can be a local server, a remote server, or a cloud server (also known as a cloud server). When the server is a non-cloud server, the non-cloud server can be connected to the cloud server for communication. The embodiments of the present application are not limited. For example, Figure 3As shown, the interview content regulation device based on multi-strategy cooperation can include a data acquisition module 301, a first strategy analysis module 302, a second strategy analysis module 303, and an interview adjustment module 304, wherein: The data acquisition module 301 is configured to acquire target interview data corresponding to the target interview content, wherein the target interview data at least includes a psychological state of the interviewee, first interview data for adjusting an interview direction and an interview stage, and second interview data for adjusting an interview action; the psychological state is obtained by comprehensively evaluating four interview parameters of the interviewee, including a trust degree, a fatigue degree, a participation degree, and an interest degree, which are calculated in advance.

[0087] The first strategy analysis module 302 is configured to perform a first strategy analysis operation on the psychological state and the first interview data according to a pre-trained first strategy analysis model, to obtain a first strategy analysis result for the psychological state and the first interview data; the first strategy analysis result includes a global control action, and the global control action includes at least one action for performing an interview stage conversion, an interview rhythm adjustment, and a topic range change.

[0088] The second strategy analysis module 303 is configured to perform a second strategy analysis operation on the psychological state and the second interview data according to a pre-trained second strategy analysis model, to obtain a second strategy analysis result for the psychological state and the second interview data; the second strategy analysis result includes a target interview action, and the target interview action includes an adjustment instruction for at least one action of a questioning manner, a response manner, a speech speed adjustment, and a tone adjustment.

[0089] The interview adjustment module 304 is configured to perform an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result, to obtain an interview adjustment result for the target interview content.

[0090] It can be seen that the implementation Figure 3 The described interview content regulation device based on multi-strategy cooperation can accurately grasp the real-time interview state of the interviewee by evaluating the psychological state of the interviewee from multiple dimensions, and can provide accurate data basis for subsequent personalized interview regulation; the double-strategy analysis model is adopted to comprehensively optimize the interview process from the macro and micro levels, which is beneficial to improving the interview quality and efficiency; at the same time, based on the strategy analysis results output by the two strategy analysis models, the interview content can be flexibly changed according to the real-time state of the interviewee, which enhances the adaptability and flexibility of the interview, and greatly improves the success rate and effect of the interview.

[0091] In an optional embodiment, the first strategy analysis model is a deep Q-learning based network model; the first interview data includes first sub-data corresponding to a dialogue progress and a cumulative feature of a dialogue history; the dialogue progress includes a shallow interview stage or a deep interview stage, and further includes a warm-up stage required or a warm-up stage not required; the cumulative feature of the history includes at least one of a historical interaction mode, a historical interview rhythm, and a historical topic type; The first strategy analysis module 302 performs the first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model, and obtains the first strategy analysis result for the psychological state and the first interview data in the following manner: According to the pre-trained first strategy analysis model, a first state vector corresponding to the psychological state and the first interview data is constructed; The first state vector is input into the first strategy analysis model, and a first Q value prediction operation is performed on the first state vector according to the first strategy analysis model, to obtain a plurality of first predicted Q values for the first state vector; each first predicted Q value corresponds to a first action, and each first action is used to adjust a direction, a stage or a range of the target interview content; According to the first strategy analysis model, a first target Q value adapted to the target interview content is selected from all the first predicted Q values in combination with a pre-set first action selection strategy; The first target Q value and the first action corresponding thereto are determined as the first strategy analysis result for the psychological state and the first interview data.

[0092] In this optional embodiment, the first state vector includes a first sub-vector corresponding to the dialogue progress and a second sub-vector corresponding to the cumulative feature of the dialogue history; The first strategy analysis module 302 performs the first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model, and obtains the first strategy analysis result for the psychological state and the first interview data in the following manner: Before the first state vector is input into the first strategy analysis model, a feature extraction and embedding operation is performed on the first sub-vector according to the first strategy analysis model, to obtain a feature extraction and embedding result for the first sub-vector; The first state vector is updated according to the feature extraction and embedding result; the feature extraction and embedding operation is used to convert the first sub-vector from a discrete vector to a continuous vector.

[0093] It can be seen that in the optional embodiment, when constructing the to-be-processed data (the first state vector), the progress of the dialogue and the historical cumulative features can be comprehensively considered, and the feature extraction and embedding conversion are performed on the related sub-vectors to convert the discrete vectors into continuous vectors that are convenient for the model to process, which is beneficial to improve the processing capability of the model. Then, the first strategy analysis model based on deep Q learning is adopted, and through the Q value prediction and selection mechanism, it is ensured that the first action output by the first strategy analysis model is adapted to the target interview content, so that based on the first strategy analysis result output by the first strategy analysis model, the interview direction, stage and range can be accurately and scientifically optimized, and the efficiency and quality of the interview are improved.

[0094] In another optional embodiment, the second strategy analysis model is a network model based on deep Q learning; the second interview data includes second sub-data corresponding to the feedback information of the last round of dialogue, the voice and the emotion analysis result; the feedback information of the last round of dialogue is used at least to indicate the feedback attitude of the interviewee for the last round of dialogue, and the feedback attitude is a positive attitude or a negative attitude; the voice and the emotion analysis result includes at least one feature of emotion, speech rate and pause corresponding to the voice of the interviewee; The first strategy analysis module 302 performs a second strategy analysis operation on the psychological state and the second interview data according to the pre-trained second strategy analysis model, and the manner of obtaining the second strategy analysis result for the psychological state and the second interview data includes: According to the pre-trained second strategy analysis model, a second state vector corresponding to the psychological state and the second interview data is constructed; The second state vector is input into the second strategy analysis model, and a second Q value prediction operation is performed on the second state vector according to the second strategy analysis model to obtain a plurality of second predicted Q values for the second state vector; each second predicted Q value corresponds to a second action, and each second action is used to adjust the interview action of the target interview content; According to the second strategy analysis model, a second target Q value adapted to the target interview content is selected from all the second predicted Q values in combination with a pre-set second action selection strategy; The second target Q value and the second action corresponding thereto are determined as the second strategy analysis result for the psychological state and the second interview data.

[0095] In this optional embodiment, the interview action includes a first sub-action corresponding to a questioning manner and a response manner, and a second sub-action corresponding to a speech rate and a tone; each first sub-action belongs to a discrete and limited action category; and the second sub-action belongs to a continuous action category or a discretizable action category.

[0096] In the optional embodiment, the questioning manner includes any one of direct questioning, guided questioning and open questioning; the responding manner includes any one of empathetic responding, feedback responding and analytical responding; and the all second predicted Q values include first sub predicted Q values corresponding to each first sub action and second sub predicted Q values corresponding to each second sub action.

[0097] In the optional embodiment, the second sub predicted Q value corresponding to each second sub action is a single predicted continuous value for speech speed / pitch, or a plurality of discrete predicted values.

[0098] It can be seen that in the optional embodiment, the required key information can be integrated when the second state vector is constructed, the data quality input into the second strategy analysis model is improved, and meanwhile, the analysis efficiency of the second strategy analysis model for the second state vector is improved. Further, the Q value prediction and selection ensure that the output second action is adapted to the target interview content, that is, the determination scientificity and accuracy of the second action are improved. In addition, each second action is intelligently subdivided according to the action category, including discrete first sub actions convenient for quick selection, and second sub actions which are continuous or can be discretized and can accurately adjust the speech speed and pitch. The scientific and refined second action is taken as the second strategy analysis result for subsequent interview adjustment, which is conducive to improving the interaction effect of the interview and the experience of the interviewee, and realizing more intelligent and efficient interview action regulation.

[0099] In yet another optional embodiment, please refer to Figure 4 , Figure 4 is another structure schematic diagram of the interview content regulation device based on multi-strategy cooperation disclosed by the embodiment of the present application. As shown in Figure 4 , the device further includes a determination module 305, a calculation module 306 and a first weighted summation module 307, wherein: The determination module 305 is configured to determine change information corresponding to the psychological state according to the interview adjustment result, wherein the change information at least includes new parameter values corresponding to the trust degree, the fatigue degree, the participation degree and the interest degree after the interview adjustment is performed; The calculation module 306 is configured to calculate parameter change information corresponding to each interview parameter according to the psychological state and the change information, wherein the parameter change information corresponding to each interview parameter includes a parameter difference value corresponding to the interview parameter and a difference value type corresponding to the parameter difference value; and the difference value type is configured to indicate whether the parameter difference value is positive or negative; The first weighted summation module 307 is configured to perform a weighted summation operation on the parameter change information corresponding to each interview parameter and a weighting coefficient set for each interview parameter, to obtain a first weighted summation result of the parameter change information corresponding to all interview parameters; and the first weighted summation result is configured to perform model parameter optimization on the first strategy analysis model.

[0100] It can be seen that in this optional embodiment, the new values of the four key parameters of the interview adjustment psychological state (trust, fatigue, participation, and interest) are accurately determined, and the scientific and quantitative parameter change information (including the difference value and the difference type) is calculated, and then the comprehensive evaluation result is obtained through weighted summation operation and used for optimizing the first strategy analysis model parameters. Through this process, the interviewee's acceptance of the interview adjustment result can be scored in a timely manner, and at the same time, it serves as auxiliary optimization data for the subsequent first strategy analysis model, that is, a timely detection mechanism for the interview adjustment result and an optimization mechanism for the first strategy analysis model are established, so that the subsequent optimized first strategy analysis model can better fit the interviewee's psychological state, thereby effectively improving the pertinence and quality of the interview and achieving more intelligent interview content regulation.

[0101] In another optional embodiment, the data acquisition module 301 is further configured to acquire new interview content fed back by the interviewee after the interview adjustment operation is performed; As Figure 4 shown, the apparatus further includes an analysis module 308 and a second weighted summation module 309, wherein: The analysis module 308 is configured to analyze the new interview content to obtain reward parameter information matched with the plurality of preset reward parameters, the reward parameter information including an emotional index corresponding to an emotional response parameter, a pause duration corresponding to a pause duration parameter, and a participation degree corresponding to an answer length and an answer depth; wherein the emotional index is used to indicate whether the interviewee is in a positive emotion or a negative emotion; the pause duration is used to indicate the feedback duration of the interviewee triggering the new interview content; and the participation degree is used to indicate the content completeness and content complexity of the new interview content. The second weighted summation module 309 is configured to perform weighted summation operation on each reward parameter information and the weighting coefficient for each reward parameter information to obtain a second weighted summation result for all the reward parameter information; and the second weighted summation result is used to optimize the model parameters of the first strategy analysis model and the second strategy analysis model, respectively.

[0102] It can be seen that in this optional embodiment, the new content fed back by the interviewee after the interview adjustment is first acquired, and then the emotional index, the pause duration, the participation degree, and other quantitative reward parameter information are analyzed in detail, and the weighted summation is performed through reasonable setting of the weighting coefficient to obtain the comprehensive evaluation result. The first and second strategy analysis models are optimized by using this result, which can comprehensively and accurately evaluate the interview adjustment effect, so that the first and second strategy analysis models are more in line with the feedback adjustment strategy of the interviewee, and the pertinence and effectiveness of the interview are improved.

[0103] Embodiment Four Please refer to Figure 5 , Figure 5is a structural schematic view of still another interview content regulation device based on multi-strategy cooperation disclosed by the embodiment of the present application. As shown in Figure 5 The interview content regulation device based on multi-strategy cooperation can include: a memory 401 storing executable program codes; a processor 402 coupled with the memory 401; The processor 402 invokes the executable program codes stored in the memory 401 to execute part or all steps of any one of the interview content regulation methods based on multi-strategy cooperation described in the embodiment one or the embodiment two of the present application.

[0104] Embodiment five The embodiment of the present application discloses a computer storage medium storing computer instructions, which, when invoked, is used to execute part or all steps of any one of the interview content regulation methods based on multi-strategy cooperation described in the embodiment one or the embodiment two of the present application.

[0105] The above-described device embodiments are only schematic, wherein the modules illustrated as separate components can or can not be physically separate, and the components illustrated as modules can or can not be physical modules, i.e., can be located in one place, or can be distributed on multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment scheme. Those skilled in the art can understand and implement without creative labor.

[0106] Those skilled in the art can clearly understand the technical solutions of the various embodiments through the above specific description of the embodiments, and the various embodiments can be realized by means of software and necessary universal hardware platforms, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in terms of contribution to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, which includes a Read-Only Memory (ROM), a Random Access Memory (RAM), a Programmable Read-only Memory (PROM), an Erasable Programmable Read Only Memory (EPROM), a One-time Programmable Read-Only Memory (OTPROM), an Electrically-Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM), or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other medium that can be used to carry or store computer readable instructions.

[0107] Finally, it should be noted that: the above-mentioned embodiments disclosed only the preferred embodiments of the present application, only for the description of the technical solutions of the present application, and not limited; although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand; it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-strategy coordination-based interview content regulation method, characterized in that, The method comprises: obtaining target interview data corresponding to target interview content, the target interview data comprising at least a psychological state of an interviewee, first interview data for adjusting an interview direction and an interview stage, and second interview data for adjusting an interview action; the psychological state is obtained by comprehensively evaluating four interview parameters of the interviewee, i.e., trustworthiness, fatigue, participation, and interest, which are calculated in advance; performing a first strategy analysis operation on the psychological state and the first interview data according to a pre-trained first strategy analysis model to obtain a first strategy analysis result for the psychological state and the first interview data; the first strategy analysis result comprises a global control action, and the global control action comprises at least one action for performing an interview stage conversion, an interview rhythm adjustment, and a topic range change; performing a second strategy analysis operation on the psychological state and the second interview data according to a pre-trained second strategy analysis model to obtain a second strategy analysis result for the psychological state and the second interview data; the second strategy analysis result comprises a target interview action, and the target interview action comprises an adjustment instruction for at least one of a questioning manner, a response manner, a speech speed adjustment, and a tone adjustment; performing an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result to obtain an interview adjustment result for the target interview content.

2. The interview content regulation method based on multi-strategy cooperation according to claim 1, characterized in that, The first strategy analysis model is a network model based on deep Q learning; the first interview data comprises first sub-data corresponding to cumulative features of a dialogue progress and a dialogue history; the dialogue progress comprises a shallow interview stage or a deep interview stage, and further comprises a warm-up stage or a non-warm-up stage; the history cumulative features comprise at least one of a history interaction mode, a history interview rhythm, and a history topic type; The first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model to obtain the first strategy analysis result for the psychological state and the first interview data comprises: constructing a first state vector corresponding to the psychological state and the first interview data according to the pre-trained first strategy analysis model; inputting the first state vector into the first strategy analysis model, and performing a first Q value prediction operation on the first state vector according to the first strategy analysis model to obtain a plurality of first predicted Q values for the first state vector; each first predicted Q value corresponds to a first action, and each first action is used to adjust an interview direction, an interview stage, or an interview range of the target interview content; selecting a first target Q value that is suitable for the target interview content from all the first predicted Q values according to the first strategy analysis model and in combination with a pre-set first action selection strategy; determining the first target Q value and the first action corresponding thereto as the first strategy analysis result for the psychological state and the first interview data. 3.The interview content regulation method based on multi-policy coordination according to claim 2, characterized in that, The first state vector comprises a first sub-vector corresponding to the progress of the dialogue, and a second sub-vector corresponding to accumulated features of the dialogue history; The method further comprises: Before inputting the first state vector into the first policy analysis model, performing feature extraction and embedding operations on the first sub-vector according to the first policy analysis model to obtain a feature extraction and embedding result for the first sub-vector; Updating the first state vector according to the feature extraction and embedding result; the feature extraction and embedding operations are used to convert the first sub-vector from a discrete vector to a continuous vector.

4. The interview content regulation method based on multi-policy coordination according to claim 1 or 2, characterized in that, The second policy analysis model is a deep Q-learning-based network model; the second interview data comprises second sub-data corresponding to last round dialogue feedback information, voice and emotion analysis results; the last round dialogue feedback information at least indicates a feedback attitude of the interviewee for the last round dialogue, and the feedback attitude is a positive attitude or a negative attitude; the voice and emotion analysis results comprise at least one feature of emotion, speech rate and pause corresponding to the voice of the interviewee; The second policy analysis result for the psychological state and the second interview data is obtained by performing a second policy analysis operation on the psychological state and the second interview data according to the pre-trained second policy analysis model, comprising: According to the pre-trained second policy analysis model, a second state vector corresponding to the psychological state and the second interview data is constructed; Inputting the second state vector into the second policy analysis model, and performing a second Q value prediction operation on the second state vector according to the second policy analysis model to obtain a plurality of second predicted Q values for the second state vector; each second predicted Q value corresponds to a second action, and each second action is used to adjust the interview action of the target interview content; According to the second policy analysis model, a second target Q value adapted to the target interview content is selected from all the second predicted Q values by combining a pre-set second action selection strategy; The second target Q value and its corresponding second action are determined as the second policy analysis result for the psychological state and the second interview data.

5. The interview content regulation method based on multi-strategy cooperation according to claim 4, characterized in that, The interview action comprises a first sub-action corresponding to a questioning manner and a response manner, and a second sub-action corresponding to a speech rate and a tone; each first sub-action belongs to a discrete and limited action category; the second sub-action belongs to a continuous action category or a discretizable action category; The questioning manner comprises any one of direct questioning, guided questioning and open questioning; the response manner comprises any one of empathetic response, feedback response and analytical response; all the second predicted Q values comprise a first sub-predicted Q value corresponding to each first sub-action and a second sub-predicted Q value corresponding to each second sub-action; The second sub-predicted Q value corresponding to each second sub-action is a single predicted continuous value for the speech rate / tone, or a plurality of discrete predicted values.

6. The interview content regulation method based on multi-policy coordination according to claim 2 or 3, characterized in that, The method further comprises: According to the interview adjustment result, determine the change information corresponding to the psychological state, the change information at least includes the new parameter value corresponding to the trust degree, fatigue degree, participation degree and interest degree of the four interview parameters after the interview adjustment is performed; According to the psychological state and the change information, calculate the parameter change information corresponding to each interview parameter, the parameter change information corresponding to each interview parameter includes the parameter difference value corresponding to the interview parameter and the difference value type corresponding to the parameter difference value; The difference value type is used to indicate that the parameter difference value is positive or negative; Perform weighted summation operation on the parameter change information corresponding to each interview parameter and the weighting coefficient set for each interview parameter to obtain the first weighted summation result of the parameter change information corresponding to all the interview parameters; The first weighted summation result is used to perform model parameter optimization on the first strategy analysis model.

7. The interview content regulation method based on multi-strategy cooperation according to claim 4, characterized in that, The method further comprises: Obtain the new interview content fed back by the interviewee after the interview adjustment operation is performed; Analyze the new interview content to obtain reward parameter information matched with multiple preset reward parameters, the reward parameter information includes emotional index corresponding to emotional response parameter, pause duration corresponding to pause duration parameter, participation degree corresponding to answer length and answer depth; wherein, the emotional index is used to indicate positive emotion or negative emotion of the interviewee; The pause duration is used to indicate the feedback duration of the interviewee triggering the new interview content; The participation degree is used to indicate the content integrity and content complexity of the new interview content; Perform weighted summation operation on each reward parameter information and the weighting coefficient of each reward parameter information to obtain the second weighted summation result of all the reward parameter information; The second weighted summation result is used to perform model parameter optimization on the first strategy analysis model and the second strategy analysis model respectively.

8. An interview content regulation device based on multi-strategy cooperation, characterized in that, The device comprises: A data acquisition module is configured to acquire target interview data corresponding to target interview content, the target interview data at least includes the psychological state of the interviewee, the first interview data for adjusting the interview direction and the interview stage, and the second interview data for adjusting the interview action; The psychological state is obtained by comprehensively evaluating the trust degree, fatigue degree, participation degree and interest degree of the interviewee which are calculated in advance; A first strategy analysis module is configured to perform first strategy analysis operation on the psychological state and the first interview data according to the pre-trained first strategy analysis model to obtain the first strategy analysis result of the psychological state and the first interview data; The first strategy analysis result includes global control action, and the global control action includes at least one action for performing interview stage conversion, interview rhythm adjustment and topic range change; a second strategy analysis module, configured to perform a second strategy analysis operation on the psychological state and the second interview data according to a pre-trained second strategy analysis model, to obtain a second strategy analysis result for the psychological state and the second interview data; the second strategy analysis result comprises a target interview action, and the target interview action comprises an adjustment instruction for at least one action of a question mode, a response mode, speed adjustment, and tone adjustment; an interview adjustment module, configured to perform an interview adjustment operation on the target interview content according to the first strategy analysis result and the second strategy analysis result, to obtain an interview adjustment result for the target interview content.

9. An interview content regulation device based on multi-strategy cooperation, characterized in that, The device comprises: a memory in which executable program codes are stored; a processor coupled with the memory; the processor invokes the executable program codes stored in the memory to execute the method for regulating interview content based on multi-strategy cooperation according to any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, and the computer instructions are invoked to execute the method for regulating interview content based on multi-strategy cooperation according to any one of claims 1-7.

Images