Data processing method and apparatus for proactively guiding a conversation

Abstract

The application discloses a data processing method and device for actively guiding a dialogue. The method comprises the following steps: obtaining user data to be processed in the application; performing continuous state-based feature extraction processing on the user data to be processed to obtain user intention feature data, which is feature data for representing a continuous state of user intention; performing action cost optimization processing on a candidate interactive action according to the user intention feature data to obtain optimized interactive action data; performing step-by-step forward deduction processing based on intention state prediction on the optimized interactive action data and the user intention feature data to obtain interactive action benefit data; screening the optimized action data corresponding to the maximum interactive action benefit data to obtain target interactive action data, and executing the target interactive action data. Through the above user intention state transition prediction and action cost optimization, the technical effect of improving the active guidance capability and the timeliness of the AI dialogue is realized.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically, to a data processing method and apparatus for proactively guiding dialogue. Background Technology

[0002] With the continuous development of artificial intelligence, AI is widely used in various industries. In human-computer interaction scenarios, AI analyzes human-computer interaction data and provides responses. In existing technologies, AI expands the user's state to include psychological states such as beliefs, goals, and emotions through data analysis of human-computer interaction. However, its actions are still mainly based on passive observation, responding after receiving the user's needs. In human-computer interaction scenarios, it lacks the ability to actively guide.

[0003] Therefore, this application is proposed to address the problem of poor proactive guidance capabilities in existing human-computer interaction AI dialogue technologies. Summary of the Invention

[0004] The main objective of this application is to provide a data processing method and apparatus for proactively guiding dialogue, in order to solve the problem of poor proactive guidance capability in existing human-computer interaction AI dialogue, and to achieve the technical effect of improving the proactive guidance capability and timeliness of AI dialogue.

[0005] To achieve the above objectives, the first aspect of this application proposes a data processing method for proactively guiding dialogue, applied to a human-computer interaction system, to enable AI to guide user dialogue in human-computer interaction scenarios, including: Acquire user data to be processed, wherein the user data to be processed is multimodal data of the user acquired by the human-computer interaction system; The user data to be processed is subjected to continuous state-based feature extraction processing to obtain user intent feature data, wherein the user intent feature data is feature data representing the continuous state of user intent; Based on the user intent feature data, the candidate interaction actions are subjected to action cost optimization processing to obtain optimized interaction action data, wherein the optimized interaction action data is used to represent the candidate interaction action data after action cost optimization. The optimized interaction action data and the user intent feature data are subjected to step-by-step forward extrapolation processing based on intent state prediction to obtain interaction action revenue data, wherein the interaction action revenue data is used to represent the revenue data corresponding to the optimized interaction action. Filter the optimized action data corresponding to the data with the greatest interactive action benefit to obtain the target interactive action data, and execute the target interactive action data.

[0006] Furthermore, the optimized interaction action data and the user intent feature data are subjected to step-by-step forward extrapolation processing based on intent state prediction to obtain interaction action benefit data, including: Acquire candidate interaction action data, and perform intent state prediction processing based on the intent state transition model on the candidate interaction action data and the user intent feature data to obtain predicted intent state feature data, wherein the predicted intent state feature data is feature data used to represent the user intent state after executing the candidate interaction action data. The predicted intent state feature data and the preset intent state data are processed based on the conversion probability to obtain the first benefit data of the interaction action, wherein the first benefit data of the interaction action is data used to represent the probability of the user's intent state being converted into the target intent state. The optimized interactive action data and the predicted intention state feature data are subjected to cost calculation processing based on step-by-step forward inference to obtain the second benefit data of the interactive action, wherein the second benefit data of the interactive action is cost data used to represent the execution of the interactive action; The interaction action revenue data is obtained based on the first revenue data and the second revenue data of the interaction action.

[0007] Furthermore, the optimized interaction action data and the user intent feature data undergo step-by-step forward extrapolation processing based on intent state prediction to obtain interaction action benefit data, including: First candidate interaction action data is obtained, and the first candidate interaction action data and the user intent feature data are subjected to intent state prediction processing based on the intent state transition model to obtain first predicted intent state feature data. The first predicted intent feature data is processed to calculate the revenue from interactive actions, resulting in the first interactive action revenue data. Acquire second candidate interaction action data, and perform intention state prediction processing based on the intention state transition model on the second candidate interaction action data and the first predicted intention state feature data to obtain second predicted intention state feature data. The second predicted intent feature data is processed to calculate the interactive action revenue, resulting in the second interactive action revenue data. The interactive action revenue data is obtained based on the first interactive action revenue data and the second interactive action revenue data.

[0008] Furthermore, the optimized interaction action data and the user intent feature data undergo step-by-step forward extrapolation processing based on intent state prediction to obtain interaction action benefit data, including: The user intent feature data is processed by similarity calculation based on preset intent state data to obtain intent state similarity data, wherein the intent state similarity data is data used to represent the degree of similarity between the user intent and the target intent state; The intent similarity data is compared with a preset intent similarity threshold. If the intent similarity data is greater than or equal to the preset intent similarity threshold, first inference step data is generated, wherein the first inference step data is data used to represent one forward inference step; If the intent similarity data is less than the preset intent similarity threshold, a second inference step data is generated, wherein the second inference step data is data used to represent multi-step forward inference; Based on the first or second deduction step data, perform forward deduction processing based on intent state prediction to obtain the interactive action benefit data.

[0009] Furthermore, it also includes: Obtain user target intent state data, wherein the user target intent state data is data used to represent the intent state that the user is expected to achieve; Based on the intention state transition model and the temporal asymmetric intervention cost function, an inverse optimal cost function is constructed. The inverse optimal cost function is used to represent the minimum cumulative cost required to reach the target intention state from any intention state. The user target intent state data is subjected to iterative solution processing based on the inverse optimal cost function to obtain inverse target interaction action data; The user intent feature data is acquired, the reverse target interaction action is executed, and the user intent state is updated until the user target intent state data is reached.

[0010] Furthermore, before acquiring the user data to be processed, the method further includes: Acquire interaction sample data, wherein the interaction sample data is relevant data used to represent human-computer interaction scenarios; The interaction sample data is processed by action extraction based on interaction state change to obtain interaction state change data, wherein the interaction state change data is data used to represent the user's intent state and interaction action when the intent state changes. The state change data is identified to obtain the intention state data before the change, the interaction action data, and the intention state data after the change. The model is trained using the pre-change intention state data and interaction action data as input and the post-change intention state data as output to obtain the intention state transition model.

[0011] Furthermore, the user data to be processed undergoes continuous state-based feature extraction processing to obtain user intent feature data, including: The user data to be processed is subjected to intent data extraction processing based on multimodal perception to obtain the user intent data to be processed; The user intent data to be processed is subjected to a first intent continuous state extraction process to obtain first intent continuous state data, wherein the first intent continuous state data is data used to represent the continuous state of user interest intensity. The user intent data to be processed is subjected to second intent continuous state extraction processing to obtain second intent continuous state data, wherein the second intent continuous state data is data used to represent the continuous state of user doubt index; The user intent data to be processed is subjected to third intent continuous state extraction processing to obtain third intent continuous state data, wherein the third intent continuous state data is data used to represent the continuous state of user trust. Intent features are extracted from the first intent continuous state data, the second intent continuous state data, and the third intent continuous state data to obtain the user intent feature data.

[0012] According to a second aspect of this application, a data processing device for actively guiding dialogue is proposed, applied to a human-computer interaction system, to realize AI-guided dialogue with users in human-computer interaction scenarios, including: The data acquisition module is used to acquire user data to be processed, wherein the user data to be processed is multimodal data of the user acquired by the human-computer interaction system; The intent feature extraction module is used to perform continuous state-based feature extraction processing on the user data to be processed to obtain user intent feature data, wherein the user intent feature data is feature data representing the continuous state of the user intent. The cost optimization module is used to perform action cost optimization processing on the candidate interaction actions based on the user intent feature data to obtain optimized interaction action data, wherein the optimized interaction action data is used to represent the candidate interaction action data after action cost optimization. The forward inference module is used to perform step-by-step forward inference processing on the optimized interaction action data and the user intent feature data based on intent state prediction to obtain interaction action benefit data, wherein the interaction action benefit data is used to represent the benefit data corresponding to the optimized interaction action. The filtering module is used to filter the optimized action data corresponding to the data with the maximum interactive action benefit, obtain the target interactive action data, and execute the target interactive action data.

[0013] According to a third aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing the computer to perform the above-described data processing method for actively guiding a dialogue.

[0014] According to a fourth aspect of this application, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the at least one processor to perform the aforementioned data processing method for actively guiding a dialogue.

[0015] The technical solutions provided by the embodiments of this application may include the following beneficial effects: In this application, user data to be processed is acquired, wherein the user data to be processed is multimodal data of the user acquired by the human-computer interaction system; feature extraction based on continuous state is performed on the user data to be processed to obtain user intent feature data, wherein the user intent feature data is feature data used to represent the continuous state of the user intent; action cost optimization processing is performed on candidate interaction actions based on the user intent feature data to obtain optimized interaction action data, wherein the optimized interaction action data is used to represent the candidate interaction action data after action cost optimization; step-by-step forward inference processing based on intent state prediction is performed on the optimized interaction action data and the user intent feature data to obtain interaction action benefit data, wherein the interaction action benefit data is used to represent the benefit data corresponding to the optimized interaction action; the optimized action data corresponding to the maximum interaction action benefit data is selected to obtain target interaction action data, and the target interaction action data is executed. Through the above user intent state transition prediction and action cost optimization, the technical effect of improving the proactive guidance capability and timeliness of AI dialogue is achieved. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 A flowchart of a data processing method for proactively guiding dialogue is provided in this application; Figure 2 A flowchart of a data processing method for proactively guiding dialogue is provided in this application; Figure 3 A flowchart of a data processing method for proactively guiding dialogue is provided in this application; Figure 4A schematic diagram of a data processing device for actively guiding dialogue provided in this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0020] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0021] Furthermore, the terms "installation," "setup," "equipped with," "connection," "linked," and "socketing" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0022] In some optional embodiments of this application, a data processing method for proactively guiding dialogue is proposed and applied to a human-computer interaction system to enable AI to guide user dialogue in human-computer interaction scenarios. Figure 1 A flowchart of a data processing method for proactively guiding dialogue is provided in this application, such as... Figure 1 As shown, the method includes the following steps: S101: Obtain user data to be processed; The user data to be processed refers to the multimodal data acquired by the human-computer interaction system from the user, including multimodal user data such as images, sound, and text. The human-computer interaction system can be used in various application scenarios, such as retail guidance, medical consultation, and educational tutoring, to enable AI-generated dialogue guidance for users in these different service scenarios.

[0023] S102: Perform continuous state-based feature extraction processing on the user data to be processed to obtain user intent feature data; User intent feature data is feature data used to represent the continuous state of user intent; In some optional embodiments of this application, a data processing method for proactively guiding a dialogue is proposed, comprising: The process involves: extracting intent data from the user data to be processed using multimodal perception to obtain user intent data; extracting a first continuous intent state from the user intent data to obtain first continuous intent state data, which represents the continuous state of user interest intensity; extracting a second continuous intent state from the user intent data to obtain second continuous intent state data, which represents the continuous state of user doubt index; extracting a third continuous intent state from the user intent data to obtain third continuous intent state data, which represents the continuous state of user trust level; and extracting intent features from the first, second, and third continuous intent state data to obtain user intent feature data.

[0024] The aforementioned multimodal perception can employ various perception methods such as semantic analysis, image recognition, and audio analysis. It provides a structured representation of the user intent data obtained from this multimodal perception, including defining user intent as a multidimensional continuous state vector, with each dimension representing: interest intensity, doubt index, price sensitivity, social motivation, time decision pressure, and trust. This forms a complete structured representation of the user's current psychological intent. In different application scenarios, continuous state vectors with different dimensional representations are extracted to achieve intent state feature extraction in scenarios such as retail guidance, medical consultation, and educational tutoring. This enables AI to proactively guide dialogue during human-computer interaction in different application scenarios.

[0025] S103: Optimize the action cost of candidate interaction actions based on user intent feature data to obtain optimized interaction action data; In existing technologies, dialogue strategy optimization methods (RLHF, DPO, etc.) all assume that the intervention cost is symmetrical in time, that is, the cost of intervention is the same at any time. However, in real-world service scenarios, the intervention cost has a high degree of temporal asymmetry. Therefore, this application optimizes the dialogue strategy and improves the timeliness of AI dialogue actions through the aforementioned action cost optimization.

[0026] Optimized interaction action data is used to represent candidate interaction action data after the action cost is optimized. Among them, the execution of the interaction action has a negative impact on the user's intention, such as leaving or refusing. The impact of the action on achieving the user's target intention defines the cost of the interaction action.

[0027] Construct the intervention cost function C( , ),in, Let t be the interactive action. Let t represent the user's intent at time t. The intervention cost is dynamically adjusted based on this intent. For example, in a retail sales scenario, the cost coefficient of sales pitches increases significantly when the user is alert or withdrawn; conversely, the cost coefficient of introductory statements decreases when the user is curious or observing. Furthermore, the parameters of this intervention cost function can be calibrated using user responses from historical interaction data to obtain the intervention cost function.

[0028] S104: Perform step-by-step forward extrapolation processing based on intent state prediction on optimized interaction action data and user intent feature data to obtain interaction action revenue data; Interaction benefit data is used to represent the benefit data corresponding to optimized interaction actions. Interaction benefit is the expected benefit obtained by comprehensively considering the probability of achieving the user's target intention state and the cost of intervention in the interaction action.

[0029] In some optional embodiments of this application, a data processing method for proactively guiding dialogue is proposed. Figure 2 A flowchart of a data processing method for proactively guiding dialogue is provided in this application, such as... Figure 2 As shown, the method includes the following steps: S201: Obtain candidate interaction action data, perform intent state prediction processing based on intent state transition model on candidate interaction action data and user intent feature data, and obtain predicted intent state feature data. Predicted intent state feature data is feature data used to represent the user's intent state after executing candidate interaction action data.

[0030] In some optional embodiments of this application, a data processing method for proactively guiding dialogue is proposed to train the aforementioned intent state transition model, including: The process involves acquiring interaction sample data, which represents relevant data in human-computer interaction scenarios; extracting actions based on interaction state changes from the interaction sample data to obtain interaction state change data, which represents the user's intent state and interaction actions when the intent state changes; identifying the state change data to obtain pre-change intent state data, interaction action data, and post-change intent state data; further, extracting the state change data based on the "intent state - verbal action - subsequent intent state" triplet; training the model with pre-change intent state data and interaction action data as input and post-change intent state data as output to obtain the intent state transition model.

[0031] In an optional embodiment of this application, an intent state transition model is constructed. The verbal actions performed by the AI ​​system at time t include attributes such as verbal type (product introduction, emotional resonance, scarcity cues, social proof, objection handling, and closing the deal), emotional tendency, and information density. P(…) is learned through a lightweight neural network. | , The parameterized approximation of the AI ​​interaction action enables the prediction of the impact of different interactive actions on the user's intention state. In this embodiment, by modeling the user's intention as a dynamically trackable multidimensional continuous state vector and constructing an intention state transition model based on the AI ​​interaction action, the active evaluation of changes in the user's intention state during the AI ​​interaction process is realized.

[0032] S202: Perform conversion probability-based calculations on the predicted intent state feature data and the preset intent state data to obtain the first benefit data of the interactive action; The first benefit data for interactive actions is data used to represent the probability of a user's intent state transforming into a target intent state; S203: Perform cost calculation processing based on step-by-step forward inference on the optimized interactive action data and predicted intention state feature data to obtain the second benefit data of the interactive action; The second benefit data for the interactive action is the cost data used to represent the cost of performing the interactive action; S204: Obtain the interaction action revenue data based on the first and second interaction action revenue data.

[0033] In some embodiments of this application, the benefits of the interactive action are:

[0034] in, For the target state, Here, N represents the cost weighting coefficient, and N represents the number of deduction steps. Given k candidate interaction actions, In the initial state The probability that a candidate interaction action will reach an approximate target state after N steps of evolution.

[0035] In some optional embodiments of this application, a data processing method for proactively guiding dialogue is proposed. Figure 3 A flowchart of a data processing method for proactively guiding dialogue is provided in this application, such as... Figure 3 As shown, the method includes the following steps: S301: Obtain first candidate interaction action data, and perform intent state prediction processing based on intent state transition model on the first candidate interaction action data and the user intent feature data to obtain first predicted intent state feature data. S302: Perform interactive action revenue calculation on the first predicted intent feature data to obtain the first interactive action revenue data; S303: Obtain the second candidate interaction action data, and perform intention state prediction processing based on the intention state transition model on the second candidate interaction action data and the first predicted intention state feature data to obtain the second predicted intention state feature data. S304: Perform interactive action revenue calculation on the second predicted intention feature data to obtain the second interactive action revenue data; S305: Obtain interactive action revenue data based on the first interactive action revenue data and the second interactive action revenue data.

[0036] In an optional embodiment of this application, when there are multi-step interactive actions, multi-step forward deduction is performed on the multi-step interactive actions to obtain the expected benefit of the interaction strategy including the multi-step interactive actions.

[0037] In some optional embodiments of this application, a data processing method for proactively guiding a dialogue is proposed, comprising: The user intent feature data is processed by similarity calculation based on preset intent state data to obtain intent state similarity data. The intent state similarity data is used to represent the degree of similarity between the user intent and the target intent state. Furthermore, the similarity between the user's current intent state vector and the target intent state vector is obtained by performing Euclidean distance calculation on the user's current intent state vector and the target intent state vector. The number of forward inference steps is determined based on the similarity. Compare data with similar intents to a preset intent similarity threshold. If the intent similarity data is greater than or equal to the preset intent similarity threshold, the first inference step data is generated, wherein the first inference step data is data used to represent one forward inference step; If the number of intent similarity data is less than the preset intent similarity threshold, a second inference step data is generated, wherein the second inference step data is used to represent multi-step forward inference; Based on the first or second simulation step data, perform forward simulation processing based on intent state prediction to obtain interactive action benefit data.

[0038] S105: Filter the optimized action data corresponding to the data with the maximum interactive action benefit, obtain the target interactive action data, and execute the target interactive action data.

[0039] The target interaction action data is the data of AI dialogue action sequences during human-computer interaction.

[0040] In another optional embodiment of this application, a data processing method for proactively guiding a dialogue is proposed, the method comprising the following steps: Acquire user target intent state data, which represents the desired intent state the user is expected to achieve; construct an inverse optimal cost function based on the intent state transition model and the temporal asymmetric intervention cost function, which represents the minimum cumulative cost required to reach the target intent state from any intent state; perform iterative solution processing on the user target intent state data based on the inverse optimal cost function to obtain inverse target interaction action data; acquire user intent feature data, execute inverse target interaction actions, update the user intent state, until the user target intent state data is reached.

[0041] In some optional embodiments of this application, multi-dimensional continuous vectors of user intent state space and target intent state are defined according to different application scenarios, thereby enabling the reuse of the same state transition modeling framework in different service scenarios such as retail shopping guides, medical consultations, and educational tutoring, and realizing proactive dialogue guidance in different application scenarios.

[0042] In another optional embodiment of this application, a data processing apparatus for proactively guiding a dialogue is provided. Figure 4 A schematic diagram of a data processing device for actively guiding dialogue provided in this application, such as... Figure 4 As shown, it includes: Data acquisition module 41 is used to acquire user data to be processed, wherein the user data to be processed is multimodal data of the user acquired by the human-computer interaction system; The intent feature extraction module 42 is used to perform continuous state-based feature extraction processing on the user data to be processed to obtain user intent feature data, wherein the user intent feature data is feature data representing the continuous state of the user intent. The cost optimization module 43 is used to perform action cost optimization processing on candidate interaction actions based on user intent feature data to obtain optimized interaction action data, wherein the optimized interaction action data is used to represent candidate interaction action data after action cost optimization. The forward inference module 44 is used to perform step-by-step forward inference processing on the optimized interaction action data and user intent feature data based on intent state prediction to obtain interaction action benefit data, wherein the interaction action benefit data is used to represent the benefit data corresponding to the optimized interaction action. The filtering module 45 is used to filter the optimized action data corresponding to the data with the greatest interactive action benefit, obtain the target interactive action data, and execute the target interactive action data.

[0043] The specific methods of execution of each unit in the above embodiments have been described in detail in the embodiments of the method, and will not be elaborated here.

[0044] In summary, this application involves: acquiring user data to be processed, wherein the user data to be processed is multimodal data of the user acquired by the human-computer interaction system; performing continuous state-based feature extraction processing on the user data to be processed to obtain user intent feature data, wherein the user intent feature data is feature data representing the continuous state of the user's intent; performing action cost optimization processing on candidate interaction actions based on the user intent feature data to obtain optimized interaction action data, wherein the optimized interaction action data is candidate interaction action data after action cost optimization; performing step-by-step forward inference processing based on intent state prediction on the optimized interaction action data and the user intent feature data to obtain interaction action benefit data, wherein the interaction action benefit data is benefit data representing the benefit data corresponding to the optimized interaction action; filtering the optimized action data corresponding to the maximum interaction action benefit data to obtain target interaction action data, and executing the target interaction action data. Through the above-mentioned user intent state transition prediction and action cost optimization, the technical effect of improving the proactive guidance capability and timeliness of AI dialogue is achieved.

[0045] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0046] Obviously, those skilled in the art should understand that the various units or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0047] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A data processing method for proactively guiding dialogue, characterized in that, Applied to human-computer interaction systems to enable AI to guide user dialogue in human-computer interaction scenarios, including: Acquire user data to be processed, wherein the user data to be processed is multimodal data of the user acquired by the human-computer interaction system; The user data to be processed is subjected to continuous state-based feature extraction processing to obtain user intent feature data, wherein the user intent feature data is feature data representing the continuous state of user intent; Based on the user intent feature data, the candidate interaction actions are subjected to action cost optimization processing to obtain optimized interaction action data, wherein the optimized interaction action data is used to represent the candidate interaction action data after action cost optimization. The optimized interaction action data and the user intent feature data are subjected to step-by-step forward extrapolation processing based on intent state prediction to obtain interaction action revenue data, wherein the interaction action revenue data is used to represent the revenue data corresponding to the optimized interaction action. Filter the optimized action data corresponding to the data with the greatest interactive action benefit to obtain the target interactive action data, and execute the target interactive action data.

2. The data processing method according to claim 1, characterized in that, The optimized interaction action data and the user intent feature data are subjected to stepwise forward extrapolation processing based on intent state prediction to obtain interaction action benefit data, including: Acquire candidate interaction action data, and perform intent state prediction processing based on the intent state transition model on the candidate interaction action data and the user intent feature data to obtain predicted intent state feature data, wherein the predicted intent state feature data is feature data used to represent the user intent state after executing the candidate interaction action data. The predicted intent state feature data and the preset intent state data are processed based on the conversion probability to obtain the first benefit data of the interaction action, wherein the first benefit data of the interaction action is data used to represent the probability of the user's intent state being converted into the target intent state. The optimized interactive action data and the predicted intention state feature data are subjected to cost calculation processing based on step-by-step forward inference to obtain the second benefit data of the interactive action, wherein the second benefit data of the interactive action is cost data used to represent the execution of the interactive action; The interaction action revenue data is obtained based on the first revenue data and the second revenue data of the interaction action.

3. The data processing method according to claim 1, characterized in that, The optimized interaction action data and the user intent feature data are subjected to stepwise forward extrapolation processing based on intent state prediction to obtain interaction action benefit data, including: First candidate interaction action data is obtained, and the first candidate interaction action data and the user intent feature data are subjected to intent state prediction processing based on the intent state transition model to obtain first predicted intent state feature data. The first predicted intent feature data is processed to calculate the revenue from interactive actions, resulting in the first interactive action revenue data. Acquire second candidate interaction action data, and perform intention state prediction processing based on the intention state transition model on the second candidate interaction action data and the first predicted intention state feature data to obtain second predicted intention state feature data. The second predicted intent feature data is processed to calculate the interactive action revenue, resulting in the second interactive action revenue data. The interactive action revenue data is obtained based on the first interactive action revenue data and the second interactive action revenue data.

4. The data processing method according to claim 1, characterized in that, The optimized interaction action data and the user intent feature data are subjected to stepwise forward extrapolation processing based on intent state prediction to obtain interaction action benefit data, including: The user intent feature data is processed by similarity calculation based on preset intent state data to obtain intent state similarity data, wherein the intent state similarity data is data used to represent the degree of similarity between the user intent and the target intent state; The intent similarity data is compared with a preset intent similarity threshold. If the intent similarity data is greater than or equal to the preset intent similarity threshold, first inference step data is generated, wherein the first inference step data is data used to represent one forward inference step; If the intent similarity data is less than the preset intent similarity threshold, a second inference step data is generated, wherein the second inference step data is data used to represent multi-step forward inference; Based on the first or second deduction step data, perform forward deduction processing based on intent state prediction to obtain the interactive action benefit data.

5. The data processing method according to claim 1, characterized in that, Also includes: Obtain user target intent state data, wherein the user target intent state data is data used to represent the intent state that the user is expected to achieve; Based on the intention state transition model and the temporal asymmetric intervention cost function, an inverse optimal cost function is constructed. The inverse optimal cost function is used to represent the minimum cumulative cost required to reach the target intention state from any intention state. The user target intent state data is subjected to iterative solution processing based on the inverse optimal cost function to obtain inverse target interaction action data; The user intent feature data is acquired, the reverse target interaction action is executed, and the user intent state is updated until the user target intent state data is reached.

6. The data processing method according to claim 1, characterized in that, Before acquiring the user data to be processed, the method further includes: Acquire interaction sample data, wherein the interaction sample data is relevant data used to represent human-computer interaction scenarios; The interaction sample data is processed by action extraction based on interaction state change to obtain interaction state change data, wherein the interaction state change data is data used to represent the user's intent state and interaction action when the intent state changes. The state change data is identified to obtain the intention state data before the change, the interaction action data, and the intention state data after the change. The model is trained using the pre-change intention state data and interaction action data as input and the post-change intention state data as output to obtain the intention state transition model.

7. The data processing method according to claim 1, characterized in that, The user data to be processed is subjected to continuous state-based feature extraction processing to obtain user intent feature data, including: The user data to be processed is subjected to intent data extraction processing based on multimodal perception to obtain the user intent data to be processed; The user intent data to be processed is subjected to a first intent continuous state extraction process to obtain first intent continuous state data, wherein the first intent continuous state data is data used to represent the continuous state of user interest intensity. The user intent data to be processed is subjected to second intent continuous state extraction processing to obtain second intent continuous state data, wherein the second intent continuous state data is data used to represent the continuous state of user doubt index; The user intent data to be processed is subjected to third intent continuous state extraction processing to obtain third intent continuous state data, wherein the third intent continuous state data is data used to represent the continuous state of user trust. Intent features are extracted from the first intent continuous state data, the second intent continuous state data, and the third intent continuous state data to obtain the user intent feature data.

8. A data processing apparatus for actively guiding dialogue, characterized in that, Applied to human-computer interaction systems to enable AI to guide user dialogue in human-computer interaction scenarios, including: The data acquisition module is used to acquire user data to be processed, wherein the user data to be processed is multimodal data of the user acquired by the human-computer interaction system; The intent feature extraction module is used to perform continuous state-based feature extraction processing on the user data to be processed to obtain user intent feature data, wherein the user intent feature data is feature data representing the continuous state of the user intent. The cost optimization module is used to perform action cost optimization processing on the candidate interaction actions based on the user intent feature data to obtain optimized interaction action data, wherein the optimized interaction action data is used to represent the candidate interaction action data after action cost optimization. The forward inference module is used to perform step-by-step forward inference processing on the optimized interaction action data and the user intent feature data based on intent state prediction to obtain interaction action benefit data, wherein the interaction action benefit data is used to represent the benefit data corresponding to the optimized interaction action. The filtering module is used to filter the optimized action data corresponding to the data with the maximum interactive action benefit, obtain the target interactive action data, and execute the target interactive action data.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the data processing method for actively guiding a dialogue as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the at least one processor to perform the data processing method for actively guiding a dialogue as described in any one of claims 1-7.

Images