A conversational topic pro-active conversational method and system

Abstract

The application relates to the field of artificial intelligence, and discloses a conversation topic active guidance conversation system, which comprises a user individuality analysis unit, a conversation management unit and a system response topic active guidance unit, the user individuality analysis unit is interconnected with the conversation management unit, the conversation management unit is interconnected with the system response topic active guidance unit; the user individuality analysis unit comprises a user portrait module and a user conversation history feature module; the conversation management unit comprises a semantic sentiment analysis module, a response strategy module and a conversation topic active guidance response generation module; and the system response topic active guidance unit comprises a system topic list module, a most relevant topic calculation and selection module and a topic guidance effect detection and response strategy adjustment module.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, specifically to a conversational topic-driven, proactively guided conversational method and system. Background Technology

[0002] Conversational systems are an important research area in artificial intelligence, with significant research implications and application value.

[0003] Conversational systems can be divided into two categories: task-guided conversational systems and non-task-guided conversational systems. Task-guided conversational systems aim to complete specific tasks based on user needs, such as helping users book ride-hailing services or hotels. Non-task-guided conversational systems interact with humans and are typically used in customer service and intelligent robots.

[0004] With the development of artificial intelligence, conversational systems can analyze user intent, understand user emotions, extract user conversation topics, and combine these with personalized factors such as conversation history to generate conversational responses that match the user's conversation topics and emotions, thereby improving user experience and satisfaction.

[0005] However, existing conversational systems emphasize the importance of user intent during conversations, focusing on a one-sided analysis and understanding of user dialogue behavior. System responses often passively match the user's topic background, neglecting the proactive guidance of the system's responses and failing to gradually guide the user's dialogue content to the topic background set by the system. They cannot provide dialogue responses that integrate system intent and user intent, and lack proactive guidance of the conversation topic. To address this, we propose a conversational method and system that proactively guides the conversation topic. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a proactively guided conversation method and system, which solves the aforementioned problems.

[0008] (II) Technical Solution

[0009] To achieve the above-mentioned objectives, the present invention provides the following technical solution: a conversation topic proactive guidance system, comprising a user personalization analysis unit, a conversation management unit, and a system response topic proactive guidance unit, wherein the user personalization analysis unit is interconnected with the conversation management unit, and the conversation management unit is interconnected with the system response topic proactive guidance unit;

[0010] The user personalization analysis unit is used to store user session history data and user profiles, and to analyze user personalization characteristics based on user session history data and user profiles.

[0011] The conversation management unit performs semantic and sentiment analysis on user input, formulates conversation response strategies, and automatically generates conversation responses that integrate proactive guidance from the system topic and user intent based on the dialogue content input by the user and the topic of the system response.

[0012] The system response topic active guidance unit is used to obtain the system topic list, calculate and select the most relevant topics, detect the system topic guidance effect, and adjust the response strategy.

[0013] The user personalization analysis unit includes a user profile module and a user session history feature module.

[0014] The conversation management unit includes a semantic sentiment analysis module, a response strategy module, and a conversation topic-based proactive response generation module;

[0015] The system's proactive topic guidance unit includes a system topic list module, a most relevant topic calculation and selection module, and a topic guidance effect detection and response strategy adjustment module.

[0016] Preferably, the topic-driven proactive conversation system AGCS is described as a quintuple:

[0017] AGCS = (In, UA, DM, AG, Out);

[0018] In represents the user input of the topic-driven proactive conversation system, and Out represents the system output of the topic-driven proactive conversation system.

[0019] A proactive conversational guidance method includes the following steps:

[0020] Step 1: Obtain the user's conversation text and the system theme background list;

[0021] Step 2: Use a natural language processing model to process the dialogue content, analyze user sentiment, topic and background, and obtain the user's topic, background and sentiment vector.

[0022] Step 3: Based on the user's emotional vector of the topic background, match it with the topic list of the proactively guided conversation system and select the most relevant system topic;

[0023] Step 4: Utilize the user personalization analysis unit to obtain user personalized characteristics based on user session history data and user profiles;

[0024] Step 5: Using the conversation management unit, select the most relevant system topics based on the user's topic background sentiment vector and user personalization characteristics, and formulate a system response strategy;

[0025] Step 6: Guided by the response strategy, use deep learning methods to process the user's topic background sentiment vector, profile and conversation history features, select the most relevant system topics, and generate system responses that integrate system topics to actively guide the response.

[0026] Step 7: Using the system response topic proactive guidance unit, based on the user dialogue content, the generated system response and the most relevant system topic, perform topic guidance effect detection, and based on the topic guidance effect detection results, execute the response strategy adjustment method and apply the new response strategy in step 5 of the next round of the conversation;

[0027] Step 8: Repeat the above steps.

[0028] Preferably, the step of obtaining user dialogue text includes, but is not limited to: obtaining the text conversation input by the user, obtaining the user's voice, and converting the voice into the user dialogue text using a speech-to-text tool.

[0029] Preferably, the second step involves inputting the user's dialogue text into a pre-trained natural language processing model to obtain the dialogue's emotion vector, and classifying the dialogue emotions into nine categories: calm, happy, fond, surprised, anxious, sad, angry, hateful, and expectant.

[0030] Based on the contextual information stored in the natural language processing model of the dialogue, combined with the results of syntactic and semantic analysis, the user's dialogue topic and background are analyzed, and a corresponding user topic-background sentiment vector is generated. The topic-background sentiment vector includes: the user's dialogue context information C, the dialogue topic T, the sentiment category E, the sentiment intensity S, and the dialogue intent I.

[0031] Preferably, the system response strategy in the fifth step includes the user topic background offset rate and the system response guidance topic background transfer rate.

[0032] Preferably, the third step includes the following:

[0033] S1: Select two features, dialogue sentiment and topic background, and calculate the feature vector of each system topic background in the system topic background list respectively;

[0034] S2: Calculate the cosine similarity between the user topic background sentiment vector and the features of each system topic background in the system topic background list;

[0035] S3: Compare cosine similarity and select the system theme background with the highest similarity to the user's theme background as the most relevant system theme.

[0036] Preferably, the method for performing the topic guidance effect detection in the seventh step is as follows:

[0037] S1: Select the topic background sentiment vector as the feature vector, and calculate the user dialogue content feature vector U, the generated system response feature vector R, and the most relevant system topic feature vector T;

[0038] S2: Calculate the cosine similarity C(U,R) between feature vectors U and R, and calculate the cosine similarity C(T,R) between feature vectors T and R.

[0039] S3: Calculate the Ct time series of user-system interactions at t, t-1, t-2, ..., tn. t (U,R), C t-1 (U,R)…C t-n (U,R) and C t (T,R), C t-1 (T,R)…C t-n (T,R);

[0040] S4: Calculate FF i =C i (U,R)-C i-1 (U,R)(i=t,t-1,...,t-(n-1)), representing the intensity of user topic background guidance within the time series, such as FF i If the value is greater than 0, the user theme background guidance strength is negative, such as FF. i If <0, the user theme background guidance strength is positive, FF i =0, then the user theme background guidance strength is 0;

[0041] S5: Calculate EE i =C i (T,R)-C i-1 (T,R)(i=t,t-1,...,t-(n-1)), characterization

[0042] The intensity of the system's thematic background guidance within the time series, such as EE i If the strength of the system's theme background guidance is >0, then the strength is positive, such as EE. i If the intensity is less than 0, the system's theme background guidance intensity is negative, such as EE. i =0, then the system's theme background guidance strength is 0;

[0043] S6: Calculate the topic guidance effect f = (N1 + N2) / n within the time series t, t-1, t-2, ..., tn, where: N1 represents the user topic background guidance intensity FF within the time series. i The number of positive and negative transitions, N2 represents the system theme background guidance intensity EE within the time series. i The number of positive and negative transitions, where n represents the length of the time series;

[0044] S7: A positive f-value for the topic guidance effect within the time series indicates that the conversation topic has a tendency to guide towards the system topic; a negative f-value for the topic guidance effect within the time series indicates that the conversation topic has a tendency to guide towards the user topic.

[0045] Preferably, the response strategy adjustment method in the seventh step is as follows:

[0046] S1: Based on the user's topic background sentiment vector, user's personalized characteristics, and the most relevant system topic, and in conjunction with topic guidance effect detection, formulate the response strategy that has the function of actively guiding the conversation topic, and guide the conversation topic.

[0047] S: Utilize the guiding role of the conversation topic to guide the background and content orientation of the conversation topic, thereby updating the system's response strategy.

[0048] (III) Beneficial Effects

[0049] Compared with the prior art, the present invention provides a method and system for proactively guiding conversations on specific topics, which has the following characteristics:

[0050] Beneficial effects:

[0051] 1. This conversation topic-guided conversation method and system incorporates proactive guidance into the system's responses, which can gradually guide the user's conversation topic to the topic set by the system, providing a conversation response that integrates the system's intent and the user's intent, thus achieving proactive guidance of the conversation topic. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0053] Figure 2 This is a schematic diagram of the overall architecture of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figure 1-2 A proactive conversational method, comprising the following steps:

[0056] S101 acquires the user's conversation text;

[0057] S102 retrieves the system theme background list;

[0058] S103 uses a natural language processing model to process dialogue content, analyze user sentiment, topic and background, and obtain user topic, background and sentiment vectors.

[0059] S104 Based on the user's topic background sentiment vector, match it with the system topic list and select the most relevant system topic;

[0060] S105 obtains personalized user characteristics based on user session history data and user profiles;

[0061] S106 Based on the user's topic background sentiment vector and the user's personalized characteristics, select the most relevant system topic and formulate a system response strategy;

[0062] S107 uses a response strategy as a guide, employs deep learning methods to process the user topic background sentiment vector, the user profile, and the conversation history features, selects the most relevant system topic, and generates a system response that integrates the system topic and actively guides the response.

[0063] S108 performs topic guidance effect detection based on the user dialogue content, the generated system response, and the most relevant system topic, and performs response strategy adjustment method based on the topic guidance effect detection result, applying the new response strategy in step S106 of the next round of conversation;

[0064] S109 Repeat steps S101 to S108 above.

[0065] Combination Figure 2 The overall architecture diagram of the conversation topic-driven proactive conversation system is as follows: A10 executes step S105 to store and extract user conversation history data and user profile features to obtain personalized user characteristics; A20 executes steps S103, S106, and S107 to process dialogue content, formulate system response strategies, and generate system responses that integrate proactive guidance of system topics; A30 executes steps S102, S104, and S108 to obtain a list of system topic backgrounds, calculate and select the most relevant system topics, perform topic guidance effect detection, and perform response strategy adjustment methods.

[0066] Combination Figure 1 Step S101, obtaining user dialogue text, includes but is not limited to: obtaining text conversations input by the user; obtaining user voice, and converting the voice into user dialogue text using a speech-to-text tool.

[0067] Combination Figure 1Step S103 uses a natural language processing model to process the dialogue content, analyze user emotions, themes and backgrounds, and obtain user theme and background emotion vectors. Specifically, it includes: inputting the user dialogue text into a pre-trained natural language processing model to obtain the dialogue emotion vectors, and classifying the dialogue emotions into nine categories, namely calm, happy, fond, surprised, anxious, sad, angry, hateful and expectant.

[0068] Based on the contextual information stored in the natural language processing model of the dialogue, combined with the results of syntactic and semantic analysis, the user's dialogue topic and background are analyzed, and a corresponding user topic-background sentiment vector is generated. The topic-background sentiment vector includes: the user's dialogue context information C, the dialogue topic T, the sentiment category E, the sentiment intensity S, and the dialogue intention I.

[0069] Combination Figure 1 Step S104 involves matching the user's topic background sentiment vector with the system topic background list to select the most relevant system topic, specifically:

[0070] Two features are selected: dialogue sentiment and topic background. The feature vector of each system topic background in the system topic background list is calculated respectively.

[0071] Calculate the cosine similarity between the user's topic background sentiment vector and the features of each system topic background in the system topic background list;

[0072] By comparing cosine similarity, the system theme background with the highest similarity to the user's theme background is selected as the most relevant system theme.

[0073] Combination Figure 1 The system response strategy described in step S106 includes, but is not limited to: user topic background offset rate and system response-guided topic background transfer rate; the response strategy is affected by the following factors:

[0074] Factor 1, the user's topic background sentiment vector as described in S103;

[0075] Factor 2, the user personalization features described in S105;

[0076] Factor 3, the most relevant system topic described in S104;

[0077] Factor 3, detection of the theme guidance effect described in S108;

[0078] screenshot Figure 1 The method for performing topic guidance effect detection based on the user dialogue content, the generated system response, and the most relevant system topic in step S108 is as follows:

[0079] Select the topic background sentiment vector as the feature vector, and calculate the user dialogue content feature vector U, the generated system response feature vector R, and the most relevant system topic feature vector T;

[0080] Calculate the cosine similarity C(U,R) between eigenvectors U and R, and calculate the cosine similarity C(T,R) between eigenvectors T and R.

[0081] Calculate Ct for user-system interaction in time series t, t-1, t-2, ..., tn. t (U,R), C t-1 (U,R)…C t-n (U,R) and C t (T,R), C t-1 (T,R)…C t-n (T,R);

[0082] Calculate FF i =C i (U,R)-C i-1 (U,R)(i=t,t-1,...,t-(n-1)), representing the intensity of user topic background guidance within the time series; such as FF i If the value is greater than 0, the user theme background guidance strength is negative, such as FF. i If <0, the user theme background guidance strength is positive, FF i =0, then the user theme background guidance strength is 0;

[0083] Calculate EE i =C i (T,R)-C i-1 (T,R)(i=t,t-1,...,t-(n-1)), characterizing the intensity of the system's thematic background guidance within the time series; such as EE i If the strength of the system's theme background guidance is >0, then the strength is positive, such as EE. i If the intensity is less than 0, the system's theme background guidance intensity is negative, such as EE. i =0, then the system's theme background guidance strength is 0;

[0084] Calculate the topic guidance effect f = (N1 + N2) / n within the time series t, t-1, t-2, ..., tn, where: N1 represents the user topic background guidance intensity FF within the time series. i The number of positive and negative transitions, N2 represents the system theme background guidance intensity EE within the time series. i The number of positive and negative transitions, where n represents the length of the time series.

[0085] A positive f-value in the time series indicates that the conversation topic tends to gravitate towards the system topic; a negative f-value in the time series indicates that the conversation topic tends to gravitate towards the user topic.

[0086] Combination Figure 1 The response strategy adjustment method described in step S108 is as follows:

[0087] Based on the user's topic background sentiment vector, the user's personalized characteristics, and the most relevant system topic, and in conjunction with the topic guidance effect detection, a response strategy (user topic background offset rate, system response guidance topic background transfer rate) with the active guidance function of the conversation topic is formulated to guide the conversation topic.

[0088] By leveraging the conversation topic guidance function, the background and content orientation of the conversation topic are guided, thereby updating the system response strategy (user topic background offset rate, system response guidance topic background shift rate).

[0089] like Figure 2 As shown, this embodiment provides a conversation topic proactive guidance system, including a user personalization analysis unit A10, a conversation management unit A20, and a system response topic proactive guidance unit A30. The user personalization analysis unit A10 is interconnected with the conversation management unit A20, and the conversation management unit A20 is interconnected with the system response topic proactive guidance unit A30.

[0090] The user personalization analysis unit A10 is used to store user session history data and user profiles, and analyze user personalization characteristics based on user session history data and user profiles.

[0091] The conversation management unit A20 performs semantic and sentiment analysis on user input, formulates conversation response strategies, and automatically generates conversation responses that integrate proactive guidance from the system topic and user intent based on the dialogue content input by the user and the topic of the system response.

[0092] The system response topic active guidance unit A30 is used to obtain the system topic list, calculate and select the most relevant topics, detect the system topic guidance effect, and adjust the response strategy.

[0093] Using equation (1), the topic-driven proactive conversation system AGCS can be described as a quintuple:

[0094] AGCS=(In, UA, DM, AG, Out) (1)

[0095] In equation (1), In represents the user input of the topic-driven proactive conversation system, and Out represents the system output of the topic-driven proactive conversation system.

[0096] The user personalization analysis unit A10 includes: a user profile module A101 and a user session history feature module A102;

[0097] The conversation management unit A20 includes a semantic sentiment analysis module A201, a response strategy module A202, and a conversation topic-based proactive response generation module A203.

[0098] The system response topic active guidance unit A30 includes a system topic list module A301, a most relevant topic calculation and selection module A302, and a topic guidance effect detection and response strategy adjustment module A303.

[0099] In this application, the conversation topic-initiated guided conversation method and system adopted in the embodiments of this application incorporate the system's proactive guidance in the system response, which can gradually guide the user's conversation topic to the topic set by the system, provide a conversation response that integrates the system's intent and the user's intent, and realize the proactive guidance of the conversation topic.

[0100] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A conversation-topic-driven, proactively guided conversation method, characterized in that, Includes the following steps: Step 1: Obtain the user's conversation text and the system theme background list; Step 2: Use a natural language processing model to process the dialogue content, analyze user sentiment, topic and background, and obtain the user's topic, background and sentiment vector. Step 3: Based on the user's emotional vector of the topic background, match it with the topic list of the proactively guided conversation system and select the most relevant system topic; Step 4: Use the user personalization analysis unit (A10) to obtain user personalization characteristics based on user session history data and user profiles; Step 5: Using the conversation management unit (A20), select the most relevant system topics based on the user's topic background sentiment vector and user personalization characteristics, and formulate a system response strategy; Step 6: Guided by the response strategy, use deep learning methods to process the user's topic background sentiment vector, profile and conversation history features, select the most relevant system topics, and generate system responses that integrate system topics to actively guide the response. Step 7: Using the system response topic active guidance unit (A30), perform topic guidance effect detection based on the user dialogue content, the generated system response and the most relevant system topic, and perform response strategy adjustment method based on the topic guidance effect detection results, and apply the new response strategy in step 5 of the next round of the conversation; The method for performing the topic guidance effect detection in the seventh step is as follows: S1: Select the topic background sentiment vector as the feature vector, and calculate the user dialogue content feature vector U, the generated system response feature vector R, and the most relevant system topic feature vector T; S2: Calculate the cosine similarity between feature vectors U and R. Calculate the cosine similarity between feature vectors T and R. ; S3: Calculate the time series of user-system interactions at t, t-1, t-2, ..., tn. , … and , … ; S4: Calculation , Characterizing the intensity of user-generated topic background guidance within the time series, such as Then the user theme background guidance strength is negative, such as The user theme background guidance strength is then positive. The user theme background guidance intensity is then 0; S5: Calculation , Characterizing the intensity of the system's thematic background guidance within the time series, such as The system's theme background guidance intensity is then positive, such as... Then the system's theme background guidance intensity is negative, such as... The system's theme background guidance intensity is then 0. S6: Calculate the topic guidance effect within the time series t, t-1, t-2, ..., tn. Where: N1 represents the intensity of user topic background guidance within the time series. The number of positive and negative transitions, N2 represents the intensity of the system's thematic background guidance within the time series. The number of positive and negative transitions, where n represents the length of the time series; S7: Time-Series Theme Guidance Effect A positive value indicates that the conversation topic has a tendency to guide the discussion towards the system topic, representing the topic guidance effect within the time series. A negative value indicates that the conversation topic tends to steer towards the user's topic.

2. The conversation topic-driven, proactively guided conversation method according to claim 1, characterized in that: The first step of obtaining user dialogue text includes: obtaining the text conversation input by the user, obtaining the user's voice, and converting the voice into the user dialogue text using a speech-to-text tool.

3. The conversation topic-driven, proactively guided conversation method according to claim 1, characterized in that: The second step involves inputting the user's dialogue text into a pre-trained natural language processing model to obtain the dialogue's emotion vector. The dialogue emotions are then categorized into nine types: calm, happy, fond, surprised, anxious, sad, angry, hateful, and expectant. Based on the contextual information stored in the natural language processing model of the dialogue, combined with the results of syntactic and semantic analysis, the user's dialogue topic and background are analyzed, and a corresponding user topic-background sentiment vector is generated. The topic-background sentiment vector includes: the user's dialogue context information C, the dialogue topic D, the sentiment category E, the sentiment intensity S, and the dialogue intent I.

4. The conversation topic-driven, proactively guided conversation method according to claim 1, characterized in that: The system response strategy in the fifth step includes the user topic background offset rate and the system response-guided topic background transfer rate.

5. The conversation topic-driven proactive conversation method according to claim 1, characterized in that: The specific content of the third step is as follows: S1: Select two features, dialogue sentiment and topic background, and calculate the feature vector of each system topic background in the system topic background list respectively; S2: Calculate the cosine similarity between the user topic background sentiment vector and the features of each system topic background in the system topic background list; S3: Compare cosine similarity and select the system theme background with the highest similarity to the user's theme background as the most relevant system theme.

6. The conversation topic-driven proactive conversation method according to claim 1, characterized in that: The method for adjusting the response strategy in the seventh step is as follows: S1: Based on the user's topic background sentiment vector, user's personalized characteristics, and the most relevant system topic, and in conjunction with topic guidance effect detection, formulate the response strategy that has the function of actively guiding the conversation topic, and guide the conversation topic. S2: Utilize the guiding role of the conversation topic to guide the background and content orientation of the conversation topic, thereby updating the system's response strategy.

7. A conversation topic-driven proactive conversation system, used to implement the conversation topic-driven proactive conversation method described in claim 1, characterized in that, It includes a user personalization analysis unit (A10), a session management unit (A20), and a system response topic proactive guidance unit (A30). The user personalization analysis unit (A10) is interconnected with the session management unit (A20), and the session management unit (A20) is interconnected with the system response topic proactive guidance unit (A30). The user personalization analysis unit (A10) is used to store user session history data and user profiles, and to analyze user personalization characteristics based on user session history data and user profiles. The conversation management unit (A20) performs semantic sentiment analysis on user input, formulates conversation response strategies, and automatically generates conversation responses that integrate proactive guidance from the system topic and user intent based on the dialogue content input by the user and the topic of the system response. The system response topic active guidance unit (A30) is used to obtain the system topic list, calculate and select the most relevant topic, detect the system topic guidance effect, and adjust the response strategy. The user personalization analysis unit (A10) includes a user profile module (A101) and a user session history feature module (A102). The conversation management unit (A20) includes a semantic sentiment analysis module (A201), a response strategy module (A202), and a conversation topic-based proactive response generation module (A203). The system response topic active guidance unit (A30) includes a system topic list module (A301), a most relevant topic calculation and selection module (A302), and a topic guidance effect detection and response strategy adjustment module (A303). The Active Guided Conversation System (AGCS) is described as a quintuple: AGCS=(In, UA, DM, AG, Out); In represents the user input of the topic-driven proactive conversation system, and Out represents the system output of the topic-driven proactive conversation system.

Images