Dialogue data processing method and device, vehicle, storage medium and program product

Abstract

The invention discloses a dialogue data processing method and device, a vehicle, a storage medium and a program product. The method comprises the steps that original dialogue data of the vehicle-mounted dialogue system are acquired, and the original dialogue data comprise user inquiry data and system reply data corresponding to the user inquiry data; determining a target type of the original dialogue data based on the user inquiry data and the system reply data; and data mining is performed on the original dialogue data based on the target type to obtain target inquiry data, and the target inquiry data is used for optimizing the vehicle-mounted dialogue system. According to the invention, the technical problem of poor dialogue response effect of the vehicle-mounted dialogue system in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving and data processing, and in particular to a method, device, vehicle, storage medium and program product for processing dialogue data. Background Art

[0002] With the continuous breakthroughs in artificial intelligence and big data technologies, in-vehicle dialogue systems are widely used in the fields of autonomous driving, car safety, smart entertainment, and vehicle intelligent control to achieve interaction between vehicles and drivers. Intelligent in-vehicle dialogue systems can improve driving safety and convenience.

[0003] However, the existing in-vehicle dialogue system relies on dialogue history data for learning and processing. The dialogue history data is usually in the form of raw data, which will result in a large amount of noise data and redundant data in the data source of dialogue processing, thereby reducing the useful information of the dialogue history data, affecting the accuracy of dialogue processing in the in-vehicle dialogue system and reducing the response ability of the in-vehicle dialogue system.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a method, device, vehicle, storage medium and program product for processing dialogue data, so as to at least solve the technical problem of poor dialogue response effect of the vehicle-mounted dialogue system in the prior art.

[0006] According to one aspect of an embodiment of the present invention, a method for processing conversation data is provided, comprising: acquiring original conversation data of an in-vehicle conversation system, wherein the original conversation data includes user inquiry data and system response data corresponding to the user inquiry data; determining a target type of the original conversation data based on the user inquiry data and the system response data; and performing data mining on the original conversation data based on the target type to obtain target inquiry data, wherein the target inquiry data is used to optimize the in-vehicle conversation system.

[0007] Optionally, based on the user inquiry data and the system response data, the target type of the original conversation data is determined, including: determining the degree of match between the user inquiry data and the system response data; when the degree of match is greater than or equal to a preset threshold, determining the target type as answered data; when the degree of match is less than a preset threshold, determining the target type as unanswered data.

[0008] Optionally, when the target type is answered data, data mining is performed on the original conversation data to obtain target inquiry data, including: aggregating the original conversation data according to the conversation time of the original conversation data to obtain two adjacent rounds of conversation data; mining user inquiry data in the previous round of conversation data based on user feedback data in the latter round of conversation data in the two adjacent rounds of conversation data to obtain first inquiry data; mining user inquiry data in the previous round of conversation data based on two user inquiry data in the two adjacent rounds of conversation data to obtain second inquiry data; merging and deduplicating the first inquiry data and the second inquiry data to obtain target inquiry data.

[0009] Optionally, based on user feedback data in the latter round of conversation data between two adjacent rounds of conversation data, user inquiry data in the previous round of conversation data is mined to obtain first inquiry data, including: inputting user feedback data and user inquiry data into a prediction model, using the prediction model to predict whether two adjacent rounds of conversation data meet user needs, and obtaining a prediction result; when the prediction result is that the two adjacent rounds of conversation data do not meet user needs, determining that the user inquiry data in the previous round of conversation data is the first inquiry data.

[0010] Optionally, based on two user inquiry data in two adjacent rounds of conversation data, the user inquiry data in the previous round of conversation data is mined to obtain second inquiry data, including: performing feature extraction on the two user inquiry data to obtain two semantic vectors; determining the target similarity of the two user inquiry data based on the text similarity between the two user inquiry data and the similarity of the two semantic vectors; when the target similarity is greater than a preset similarity, determining the user inquiry data in the previous round of conversation data as the second inquiry data.

[0011] Optionally, based on the text similarity between the two user query data and the similarity of the two semantic vectors, the target similarity of the two user query data is determined, including: determining the text similarity between the two user query data to obtain a first similarity result; determining the similarity of the two semantic vectors to obtain a second similarity result; and weighting the first similarity result and the second similarity result to obtain the target similarity.

[0012] Optionally, when the target type is unanswered data, data mining is performed on the original conversation data to obtain target inquiry data, including: filtering the noise data in the original conversation data to obtain first filtered data; filtering the conversation data irrelevant to the vehicle-mounted conversation system in the first filtered data to obtain second filtered data; and determining the target inquiry data based on the initial inquiry data in the second filtered data.

[0013] Optionally, filtering the noise data in the original conversation data to obtain first filtered data includes one of the following: performing noise detection on the original conversation data using a first language model to obtain a detection result of the original conversation data, and filtering the noise data in the original conversation data based on the detection result to obtain the first filtered data; clustering the original conversation data using a density clustering algorithm to obtain a clustering result, and deleting other conversation data in the original conversation data except the clustering result to obtain the first filtered data.

[0014] Optionally, filtering the conversation data in the first filtered data that is not related to the vehicle-mounted dialogue system to obtain second filtered data includes: using a second language model to predict the type of the first filtered data to obtain the type of the first filtered data, wherein the type is used to characterize whether the first filtered data is related to the vehicle-mounted dialogue system; filtering the conversation data in the first filtered data that is not related to the vehicle-mounted dialogue system based on the type of the first filtered data to obtain the second filtered data.

[0015] Optionally, based on the initial query data in the second filtered data, target query data is determined, including: vectorizing the initial query data to obtain a query semantic vector; converting the query semantic vector to a three-dimensional data space to obtain a converted semantic vector; performing data aggregation based on the converted semantic vector to obtain at least one cluster center; and determining the initial query data corresponding to at least one cluster center as the target query data.

[0016] Optionally, the above method also includes: associating the target inquiry data with the knowledge data of the in-vehicle dialogue system to obtain associated knowledge data; optimizing the in-vehicle dialogue system based on the target training data and the associated knowledge data, wherein the target training data is different from the training data of the in-vehicle dialogue system.

[0017] According to another aspect of an embodiment of the present invention, a device for processing conversation data is also provided, including: an acquisition module, used to acquire original conversation data of an in-vehicle conversation system, wherein the original conversation data includes user inquiry data and system response data corresponding to the user inquiry data; a determination module, used to determine a target type of the original conversation data based on the user inquiry data and the system response data; and a mining module, used to perform data mining on the original conversation data based on the target type to obtain target inquiry data, wherein the target inquiry data is used to optimize the in-vehicle conversation system.

[0018] According to another aspect of an embodiment of the present invention, a vehicle is further provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the method in each embodiment of the present invention is executed when the program is running.

[0019] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0020] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0021] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0022] According to another aspect of the embodiments of the present invention, a computer program is further provided. When the computer program is executed by a processor, the methods in the embodiments of the present invention are implemented.

[0023] In an embodiment of the present invention, original dialogue data of an in-vehicle dialogue system is obtained, wherein the original dialogue data includes user inquiry data and system response data corresponding to the user inquiry data; based on the user inquiry data and the system response data, the target type of the original dialogue data is determined; based on the target type, data mining is performed on the original dialogue data to obtain target inquiry data, wherein the target inquiry data is used to optimize the in-vehicle dialogue system. It is easy to notice that by using data mining, different target types of the original dialogue data are distinguished, and different data mining methods are used for original dialogue data of different target types to obtain target inquiry data, so that the user's intention can be mined and understood, thereby more accurately understanding the user's real needs, and because the target inquiry data is obtained by data mining the original dialogue data of the in-vehicle dialogue system, it can more accurately reflect the insufficiency of the dialogue response of the in-vehicle dialogue system, so as to optimize the in-vehicle dialogue system by using the target inquiry data, and achieve the purpose of effective response, thereby achieving the technical effect of improving the dialogue response capability of the in-vehicle dialogue system, thereby solving the technical problem of poor dialogue response effect of the in-vehicle dialogue system in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0025] Figure 1 is a schematic diagram of a method for processing conversation data according to an embodiment of the present invention;

[0026] Figure 2 is a flow chart of a method for processing conversation data according to an embodiment of the present invention;

[0027] Figure 3 is a schematic diagram of a target type of original conversation data according to an embodiment of the present invention;

[0028] Figure 4 is a schematic diagram of a prediction model according to an embodiment of the present invention;

[0029] Figure 5 is a schematic diagram of an optional method for obtaining target query data according to an embodiment of the present invention;

[0030] Figure 6 is a schematic diagram of a clustering result according to an embodiment of the present invention;

[0031] Figure 7 is a schematic diagram of data aggregation according to an embodiment of the present invention;

[0032] Figure 8 is a schematic diagram of an optional method for obtaining target query data according to an embodiment of the present invention;

[0033] Fig. 9 is a schematic diagram of an optional method for processing conversation data according to an embodiment of the present invention;

[0034] Fig.10 2 is a schematic diagram of a device for processing conversation data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0037] According to an embodiment of the present invention, a method embodiment of a method for processing conversation data is provided. It should be noted that the steps shown in the flowchart of 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 can be executed in an order different from that shown here.

[0038] The embodiment of the present application provides a method for processing dialogue data. The method for processing dialogue data can be used to provide a dialogue function for an in-vehicle dialogue system. The preset application scenarios of the above-mentioned in-vehicle dialogue system may include the following scenarios in the vehicle field: commuting automatic driving scenarios, artificial intelligence (AI) driving scenarios for family cars, automatic parking assistance (APA) scenarios (such as memory parking for owned parking spaces in garages, smart parking for designated parking spaces in parking lots, etc.), and intelligent navigation assistance (Navigation Guided Pilot, NGP) scenarios in urban areas or high-speed areas. In addition, the above-mentioned preset application scenarios may also include, but are not limited to: dialogue scenarios of smart driving trucks or unmanned trucks in the field of logistics and transportation, dialogue scenarios of self-driving agricultural vehicles in the field of agricultural machinery, dialogue scenarios of drones, and dialogue scenarios of intelligent robots (such as cleaning robots, service robots, delivery robots, etc.).

[0039] When the above-mentioned preset application scenarios are scenarios in fields other than the vehicle field, those skilled in the art should be able to understand that the vehicle in the above-mentioned method for processing dialogue data can be replaced with other objects (such as agricultural machinery, drones, robots, etc.), and accordingly, the vehicle-mounted dialogue system can be replaced with a dialogue system related to other objects. On this basis, in the embodiment of the present application, the specific implementation method of the above-mentioned method for processing dialogue data is exemplified by taking the field of intelligent driving as an example.

[0040] As an optional implementation, the above-mentioned method for processing conversation data can be applied to, but not limited to, Figure 1 The application scenario shown. Figure 1 is a schematic diagram of an application scenario of a method for processing conversation data according to an embodiment of the present invention, such as Figure 1 As shown, in the application scenario, the terminal device 10 may, but is not limited to, communicate with the server 14 through the network 12, and the server 14 may, but is not limited to, perform operations on the database, such as write data operations or read data operations. The above-mentioned terminal device 10 may include, but is not limited to, a human-computer interaction screen, a processor, and a memory. The above-mentioned vehicle 16 may, but is not limited to, be used to respond to the above-mentioned human-computer interaction operations, perform corresponding operations, or generate corresponding instructions and send the generated instructions to the server 14. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in an order different from that here.

[0041] Figure 2 is a flow chart of a method for processing conversation data according to an embodiment of the present invention. Figure 2 As shown, the method comprises the following steps:

[0042] Step S202, obtaining original dialogue data of the vehicle-mounted dialogue system.

[0043] The original conversation data includes user inquiry data and system response data corresponding to the user inquiry data.

[0044] The above-mentioned in-vehicle dialogue system can be an intelligent voice interaction system with autonomous learning capabilities. The in-vehicle dialogue system uses technologies such as natural language processing, speech recognition, and speech synthesis to enable the driver or passenger to interact with the vehicle through voice commands, thereby performing various operations or obtaining information without manual operation by the driver or passenger. Through the in-vehicle dialogue system, driving safety, convenience, and comfort can be improved, and the driver's manual operation during driving can be reduced. At the same time, a rich variety of information and entertainment services are provided to provide users with more intelligent driving assistance. The functions of the in-vehicle dialogue system may include but are not limited to: voice control, information query, entertainment functions, schedule management, vehicle status monitoring, personalized services, and emergency rescue. They are not limited here and can be determined as needed. The in-vehicle dialogue system can dynamically capture and understand user intentions through real-time analysis and deep mining of user interaction data, realize self-expansion and optimization of the knowledge base, thereby improving the intelligence level and service efficiency of the in-vehicle dialogue system.

[0045] The above-mentioned raw dialogue data may include user inquiry data and system response data corresponding to the user inquiry data. The raw dialogue data may be unprocessed voice or text data directly derived from the user's interaction with the in-vehicle dialogue system. In the in-vehicle dialogue system, the raw dialogue data may include text data corresponding to the voice commands issued by the user through the in-vehicle microphone, and the feedback of the in-vehicle dialogue system based on the voice commands. The raw dialogue data is the basis for the in-vehicle dialogue system to train and optimize the model. By learning from the raw dialogue data, the in-vehicle dialogue system can understand the user's intentions and improve the accuracy of the response and the fluency of the dialogue.

[0046] The above-mentioned user inquiry data may refer to questions, requests or instructions raised by the user to the system in voice or text form during the use of the intelligent dialogue system. User inquiry data is the basis for the interaction of the in-vehicle dialogue system, and it is also the starting point for the in-vehicle dialogue system to understand and respond to user needs. In the in-vehicle dialogue system, for the in-vehicle dialogue system with voice input, the user inquiry data may be the user's voice command captured by the microphone inside the vehicle, and then converted into text form for processing by the in-vehicle dialogue system. For the in-vehicle dialogue system with text input, the user inquiry data may be a directly input text message. By analyzing the user inquiry data, the in-vehicle dialogue system can understand the user's preferences and usage habits, thereby improving the knowledge base, enhancing the rationality of the dialogue process design, and enhancing the intelligent and personalized service capabilities of the in-vehicle dialogue system.

[0047] The above-mentioned system response data can be used to characterize the response of the in-vehicle dialogue system to the user's inquiry. The system response data can be divided into the following types: direct response data of the in-vehicle dialogue system, clarified response data of the in-vehicle dialogue system, and backup data of the in-vehicle dialogue system. Among them, the direct response data can be a direct and clear response given immediately when the in-vehicle dialogue system can directly understand the intention of the user's query and find matching solutions or information. The clarified response data can be when the in-vehicle dialogue system receives ambiguous or insufficient information from the user's inquiry data, and asks the user in a clarifying way to confirm the user's true intention through further interaction with the user. The backup data usually provides a default and universal response to the user's inquiry that the in-vehicle dialogue system cannot recognize or directly respond to, avoiding complete silence or erroneous output. In the in-vehicle dialogue system, the system response data is usually played to the user in the form of voice through the vehicle's audio system. The quality of the system response data directly affects the user experience and the efficiency of the in-vehicle dialogue system.

[0048] In an optional embodiment, the original conversation data can be obtained through a log file. Specifically, the sound in the cockpit can be monitored through the vehicle microphone. When the voice command issued by the user is detected, the user inquiry data recorded in real time by the vehicle microphone is generated into a log file for subsequent data processing. This process ensures the real-time and integrity of the original conversation data. Then, the system response data of the vehicle dialogue system to the user inquiry data can also be recorded in the log file, so that the log file is saved so that the original conversation data can be obtained through the log file later.

[0049] In another optional embodiment, the original conversation data can be obtained through the cloud platform. After the vehicle is used, the original conversation data stored in the local log file will be uploaded to the cloud platform. These conversation data may contain conversation records of multiple users in different time periods. The cloud platform archives and stores the uploaded original conversation data, and can also classify, clean and pre-process it for subsequent analysis and mining.

[0050] Step S204: determining the target type of the original conversation data based on the user inquiry data and the system response data.

[0051] Distinguish the target type so that different data mining methods can be used for different raw dialogue data to obtain accurate target inquiry data. By analyzing user inquiry data and system response data, it is possible to identify dialogue scenarios that the in-vehicle dialogue system fails to properly handle, including but not limited to system response errors, incomplete responses, no responses, etc. Distinguishing the target type of the original dialogue data helps to clarify the problems and optimization directions of the in-vehicle dialogue system. In addition, the target type can provide feedback information for the in-vehicle dialogue system, so that the in-vehicle dialogue system can improve its knowledge base and algorithm based on actual user interaction data. For example, for scenarios where the in-vehicle dialogue system responds incorrectly, the relevant entries in the knowledge base can be updated; for the underlying data that cannot be identified, the training of the natural language processing model can be improved, thereby improving the system's understanding and responsiveness.

[0052] In the interactive process of the in-vehicle dialogue system, the user inquiry data and the system response data constitute a complete dialogue record. The original dialogue data can be divided into answered data and unanswered data according to whether the in-vehicle dialogue system successfully generates and sends a response that meets the user's needs. Answered data can refer to the data that the in-vehicle dialogue system successfully identifies the user's intention and generates a corresponding response after receiving the user's inquiry data. The answered data reflects the dialogue effectiveness and response quality of the in-vehicle dialogue system. Unanswered data can refer to the data that the in-vehicle dialogue system fails to generate a response that meets the user's needs after receiving the user's inquiry data, or generates a response but the user clearly expresses dissatisfaction. By mining the potential information in these two types of data, the intelligence level of the in-vehicle dialogue system and its ability to cope with complex user needs can be improved. In addition, unanswered data is an object that needs attention and optimization. Through unanswered data, the current limitations and improvement points of the in-vehicle dialogue system can be revealed. By mining and analyzing unanswered data, the in-vehicle dialogue system can discover gaps in the knowledge base, defects in the algorithm, and possible changes in user needs, so as to carry out targeted updates and optimizations to improve service accuracy and user satisfaction.

[0053] In an optional embodiment, the target type of the original dialogue data can be determined according to a classification model. Specifically, the user inquiry data and the system response data can be input into the classification model, and the classification model can classify the original dialogue data and output the target type. The classification model can be one of the following: support vector machine, decision tree, random forest, deep learning model, which is not limited here and can be determined as needed. Among them, the initial classification model can use the labeled user inquiry data and system response data as a training set, and optimize by adjusting the parameters of the initial classification model and using cross-validation, etc., to ensure that the trained classification model can accurately distinguish between answered data and unanswered data. Therefore, the trained classification model can be used to predict and mark whether the original dialogue data is answered data or unanswered data, and the target type of the original dialogue data can be determined.

[0054] In another optional embodiment, the target type of the original dialogue data can be distinguished according to the matching degree between the user inquiry data and the system response data. Specifically, when the matching degree is greater than or equal to the preset threshold, the target type is determined to be answered data, and when the matching degree is less than the preset threshold, the target type is determined to be unanswered data. Alternatively, the target type can also be determined according to the user's satisfaction with the system response data, so as to subsequently mine the target inquiry data of unmet needs reported by the user, thereby improving the response capability of the in-vehicle dialogue system. Alternatively, the target type can also be determined according to the dialogue turn.

[0055] Step S206: performing data mining on the original dialogue data based on the target type to obtain target inquiry data.

[0056] Among them, the target inquiry data is used to optimize the in-vehicle dialogue system.

[0057] The above-mentioned data mining refers to the process of analyzing and processing the original conversation data. According to the target type, different data mining methods are used on the original conversation data to improve the applicability of data mining.

[0058] The above-mentioned target inquiry data can be used to optimize the in-vehicle dialogue system. The target inquiry data can be obtained by data mining the original dialogue data in the in-vehicle dialogue system. The target inquiry data can be user inquiry data that the in-vehicle dialogue system fails to correctly identify, understand, or fails to provide an effective response. For example, user inquiry data that gives an incorrect response, an incomplete response, or the user clearly expresses dissatisfaction with the response. The target inquiry data can also be blank points in the knowledge base of the in-vehicle dialogue system, and user inquiry data that cannot provide a response. The target inquiry data can also be high-value user inquiry data to provide a guarantee for the accuracy of the in-vehicle dialogue system. The target inquiry data is not limited here and can be determined as needed. Through the target inquiry data, the knowledge base content can be optimized, missing knowledge points can be supplemented, erroneous information can be corrected, and the knowledge base can be made more comprehensive and accurate. It can also enhance the in-vehicle dialogue system's ability to understand complex contexts and user intentions, and improve the fluency and efficiency of the dialogue.

[0059] In an optional embodiment, data mining can be performed on the original dialogue data based on the target type to obtain the target inquiry data, so as to perform data mining on different original dialogue data, thereby optimizing the vehicle dialogue system using the target inquiry data. Optionally, data mining can be further performed on the answered data in the original dialogue data. For example, the user feedback data in the original dialogue data can be identified by the recognition model, and the user feedback data can be further used as the basis for optimizing the vehicle dialogue system to improve the adaptability of the vehicle dialogue system. Alternatively, the important data in the original dialogue data can be analyzed by cluster analysis, thereby optimizing the vehicle dialogue system. In addition, data mining can be further performed on the unanswered data in the original dialogue data. For example, the noise data can be detected by the first detection model to continue to process the noise data to ensure the accuracy of the data processed by the vehicle dialogue system, or the irrelevant data can be detected by the second detection model to filter the irrelevant data, thereby ensuring the relevance of the target inquiry data and the vehicle dialogue system. Alternatively, visualization and cluster analysis can be performed on the unanswered data to ensure the relevance of the objects of subsequent data mining.

[0060] In an embodiment of the present invention, original dialogue data of an in-vehicle dialogue system is obtained, wherein the original dialogue data includes user inquiry data and system response data corresponding to the user inquiry data; based on the user inquiry data and the system response data, the target type of the original dialogue data is determined; based on the target type, data mining is performed on the original dialogue data to obtain target inquiry data, wherein the target inquiry data is used to optimize the in-vehicle dialogue system. It is easy to notice that by using data mining, different target types of the original dialogue data are distinguished, and different data mining methods are used for original dialogue data of different target types to obtain target inquiry data, so that the user's intention can be mined and understood, thereby more accurately understanding the user's real needs, and because the target inquiry data is obtained by data mining the original dialogue data of the in-vehicle dialogue system, it can more accurately reflect the insufficiency of the dialogue response of the in-vehicle dialogue system, so as to optimize the in-vehicle dialogue system by using the target inquiry data, and achieve the purpose of effective response, thereby achieving the technical effect of improving the dialogue response capability of the in-vehicle dialogue system, thereby solving the technical problem of poor dialogue response effect of the in-vehicle dialogue system in the prior art.

[0061] Optionally, based on the user inquiry data and the system response data, the target type of the original conversation data is determined, including: determining the degree of match between the user inquiry data and the system response data; when the degree of match is greater than or equal to a preset threshold, determining the target type as answered data; when the degree of match is less than a preset threshold, determining the target type as unanswered data.

[0062] The above-mentioned matching degree can be a display threshold, and the matching degree is used to preliminarily screen and classify the unanswered data in the data mining process. In the scenario of the in-vehicle dialogue system, users may make a variety of different requests, and it is necessary to select appropriate or highly important ones for deep mining and model training. Therefore, by setting the display threshold, the unanswered data can be preliminarily roughly classified to determine the original dialogue data worthy of further analysis.

[0063] The display threshold can be a boundary value in the raw dialogue data that needs to be further processed. The display threshold can be used to screen, filter or sort data. By setting a reasonable display threshold, it can be used to filter out raw dialogue data below the display threshold, and it can also be used to filter high-value raw dialogue data, exclude noise or low-relevance raw dialogue data, and ensure that the raw dialogue data used for knowledge base update or algorithm optimization is of high quality. The setting of the display threshold depends on user needs. For example, if the focus of the in-vehicle dialogue system is to improve the accuracy of the response, then a higher display threshold needs to be set; if the in-vehicle dialogue system is more concerned with covering the breadth of user queries, then a lower display threshold may be set lower to include more relevant raw dialogue data. Adjusting the display threshold can be dynamically optimized based on the dialogue effect and user feedback. Too high or too low a display threshold may affect the user experience and the response performance of the in-vehicle dialogue system.

[0064] by Figure 3 Describes the target type of the raw conversation data, Figure 3 is a schematic diagram of a target type of original conversation data according to an embodiment of the present invention, such as Figure 3 As shown, the dialogue mechanism of the in-vehicle dialogue system can be obtained from the online log. The dialogue mechanism includes direct, clarification, and backup. According to different dialogue mechanisms, the target type of the original dialogue data is divided into answered data and unanswered data.

[0065] Optionally, when the target type is answered data, data mining is performed on the original conversation data to obtain target inquiry data, including: aggregating the original conversation data according to the conversation time of the original conversation data to obtain two adjacent rounds of conversation data; mining user inquiry data in the previous round of conversation data based on user feedback data in the latter round of conversation data in the two adjacent rounds of conversation data to obtain first inquiry data; mining user inquiry data in the previous round of conversation data based on two user inquiry data in the two adjacent rounds of conversation data to obtain second inquiry data; merging and deduplicating the first inquiry data and the second inquiry data to obtain target inquiry data.

[0066] In an optional embodiment, the context dialogue may be determined based on time, and then the target query data may be mined from the context dialogue.

[0067] Specifically, two adjacent rounds of dialogue data may be determined by dialogue time. Optionally, dialogue data with a difference less than a preset difference may be determined as adjacent rounds of dialogue by the difference of dialogue time, thereby determining adjacent rounds of dialogue data. Alternatively, dialogue time may be clustered by cluster analysis to obtain adjacent rounds of dialogue, thereby determining adjacent rounds of dialogue data corresponding to the adjacent rounds of dialogue. Alternatively, adjacent rounds of dialogue data may be determined by aggregation of a time window of dialogue time. First, a time window is defined for aggregating adjacent rounds of dialogue data, such as 10 seconds, 15 seconds, etc., which is not limited here and can be determined as needed. Based on the time window, the original dialogue data is aggregated in time series. For example, for each user inquiry data, check whether there is another user inquiry data or system response data within the next 10 seconds. If so, the two user inquiry data are regarded as a group of adjacent rounds of dialogue data. If no user inquiry data or system response data is found within the time window, the record is regarded as a separate dialogue data and does not participate in the aggregation of adjacent dialogues. Then, high-quality data is mined, including user active feedback data, such as user feedback replies such as "wrong", "wrong", etc., and contextual semantic mining. Furthermore, the two mining results are summarized and deduplicated. Through this step, it is possible to quickly identify which original dialogue data has been successfully processed and which original dialogue data needs further optimization, thereby improving the response ability and user satisfaction of the in-vehicle dialogue system in a targeted manner. In this way, it is easy to learn from the real feedback of users and continuously optimize the in-vehicle dialogue system's understanding of user intentions, especially in speech recognition and natural language processing, making the in-vehicle dialogue system more intelligent and better able to adapt to the language habits and needs of different users.

[0068] Optionally, based on user feedback data in the latter round of conversation data between two adjacent rounds of conversation data, user inquiry data in the previous round of conversation data is mined to obtain first inquiry data, including: inputting user feedback data and user inquiry data into a prediction model, using the prediction model to predict whether two adjacent rounds of conversation data meet user needs, and obtaining a prediction result; when the prediction result is that the two adjacent rounds of conversation data do not meet user needs, determining that the user inquiry data in the previous round of conversation data is the first inquiry data.

[0069] The above-mentioned prediction model can be a model for predicting whether two adjacent rounds of dialogue data meet the user's needs. The prediction model can be a long short-term memory network, a gated recurrent unit, or a bidirectional transformer encoder, which is not limited here and can be determined as needed. Through the prediction model, the in-vehicle dialogue system can more accurately determine whether its answer meets the user's needs, so that in subsequent optimization, it can focus more on improving the user experience, such as music playback, weather query, route planning, etc.

[0070] The prediction model can also adopt the architecture of encoder and decoder, such as Figure 4 As shown, the prediction model includes an encoder and a decoder, and the user feedback data and the user query data are input into the prediction model, and the context vector is formed by the encoder. The decoder can output the prediction result based on the context vector.

[0071] Specifically, the encoder can process the user feedback data and the user inquiry data, extract semantic features, and convert them into vector representations. The decoder can output a prediction result based on the context vector. The prediction result can characterize whether the two adjacent rounds of dialogue data meet the user's needs. When the prediction result is that the two adjacent rounds of dialogue data do not meet the user's needs, the user inquiry data in the previous round of dialogue data is determined to be the first inquiry data, that is, the user inquiry data needs to be further analyzed to improve the response capability of the in-vehicle dialogue system.

[0072] Optionally, based on two user inquiry data in two adjacent rounds of conversation data, the user inquiry data in the previous round of conversation data is mined to obtain second inquiry data, including: performing feature extraction on the two user inquiry data to obtain two semantic vectors; determining the target similarity of the two user inquiry data based on the text similarity between the two user inquiry data and the similarity of the two semantic vectors; when the target similarity is greater than a preset similarity, determining the user inquiry data in the previous round of conversation data as the second inquiry data.

[0073] In an optional embodiment, the features of the two user query data may be first extracted to obtain two semantic vectors, and then the similarity of the two semantic vectors may be calculated to identify the similarity of the two user query data in the semantic space. The text of the user query data may be matched to calculate the text similarity of the two user query data. Based on the semantic vector similarity and text similarity of the two user query data, contextual semantic mining may be performed more accurately. In this way, the in-vehicle dialogue system may be helped to identify and understand the user's possible repeated inquiries or similar needs, thereby optimizing the response strategy of the in-vehicle dialogue system, reducing the user's waiting time, and improving the safety during driving.

[0074] Optionally, based on the text similarity between the two user query data and the similarity of the two semantic vectors, the target similarity of the two user query data is determined, including: determining the text similarity between the two user query data to obtain a first similarity result; determining the similarity of the two semantic vectors to obtain a second similarity result; and weighting the first similarity result and the second similarity result to obtain the target similarity.

[0075] In an optional embodiment, the similarity of the text between the two user query data can be calculated by a text similarity algorithm. For example, the similarity of the text between the two user query data can be calculated by cosine similarity, Jaccard similarity, and semantic role labeling similarity. Then, the similarity of the two semantic vectors can be determined by Euclidean distance, semantic hashing, K-nearest neighbor algorithm, etc. to obtain a second similarity result. Then, the first similarity result and the second similarity result are weighted respectively to obtain the target similarity. Through weighted processing, the in-vehicle dialogue system can more comprehensively evaluate the similarity of user queries, which is particularly important when processing user queries in a multilingual environment, and can ensure that the in-vehicle dialogue system can accurately understand user intentions in different languages ​​and dialects.

[0076] Optionally, the target similarity can be calculated using the following formula:

[0077] S=j(w)*0.13+(2*ks(v)-1)*0.87;

[0078] Where S is the target similarity, j(w) is the text similarity calculated by the Jaccard algorithm, w is the text between the two user query data, ks(v) is the K nearest neighbor algorithm score for calculating the two semantic vectors, v is the two semantic vectors, 0.13 and 0.87 are the weights of text similarity and semantic vector similarity respectively, and * is a multiplication operation.

[0079] by Figure 5 The following describes the process of obtaining target query data when the target type is answered data. Figure 5 is a schematic diagram of an optional method for obtaining target query data according to an embodiment of the present invention, such as Figure 5 As shown, the original conversation data is aggregated to obtain two adjacent rounds of conversation data, and further, the user feedback data and two user inquiry data in the two adjacent rounds of conversation data are merged and deduplicated to obtain the target inquiry data. The user feedback data is the user feedback data in the latter round of conversation data in the two adjacent rounds of conversation data.

[0080] Optionally, when the target type is unanswered data, data mining is performed on the original conversation data to obtain target inquiry data, including: filtering the noise data in the original conversation data to obtain first filtered data; filtering the conversation data irrelevant to the vehicle-mounted conversation system in the first filtered data to obtain second filtered data; and determining the target inquiry data based on the initial inquiry data in the second filtered data.

[0081] In an optional embodiment, for the target type of unanswered data, data mining can be performed on the original conversation data by filtering out noise data and irrelevant conversation data to obtain the user's real needs, and mining can be performed on the filtered original conversation data. By effectively removing invalid conversations, the focus is on user inquiries that the in-vehicle dialogue system fails to correctly answer, thereby achieving maximum optimization of the in-vehicle dialogue system under limited resources, especially when dealing with user inquiries in emergency situations, and being able to quickly identify and improve the response of the in-vehicle dialogue system.

[0082] Optionally, filtering the noise data in the original conversation data to obtain first filtered data includes one of the following: performing noise detection on the original conversation data using a first language model to obtain a detection result of the original conversation data, and filtering the noise data in the original conversation data based on the detection result to obtain the first filtered data; clustering the original conversation data using a density clustering algorithm to obtain a clustering result, and deleting other conversation data in the original conversation data except the clustering result to obtain the first filtered data.

[0083] The first language model can be used to detect abnormal conversations in the original conversation data to filter out abnormal conversations. The first language model can be one of the following: a recurrent neural network, a long short-term memory network, etc., which is not limited here and can be determined as needed.

[0084] In an optional embodiment, the first language model can be used for preliminary filtering to exclude abnormal data with too high a score. For example, the abnormal data can be data with grammatical errors or spelling errors, random strings or meaningless data: such as "Haha", "!", etc., or irrelevant or inappropriate data. Alternatively, a density clustering algorithm can be used for clustering, and data with low density and far away from the cluster center will be identified as noise data and deleted from the original conversation data. Through the first language model and the density clustering algorithm, the in-vehicle dialogue system can effectively identify and filter out irrelevant or meaningless conversation data, improve the efficiency and quality of data mining, and especially when processing a large amount of user data, can quickly locate key information and improve the response speed and accuracy of the in-vehicle dialogue system.

[0085] Optionally, the original conversation data is clustered using a density clustering algorithm, and the clustering results are as follows: Figure 6 As shown, in Figure 6 The points of different sizes represent different original conversation data, and the points enclosed by the rectangular box are noise data, which need to be deleted from the original conversation data.

[0086] Optionally, filtering the conversation data in the first filtered data that is not related to the vehicle-mounted dialogue system to obtain second filtered data includes: using a second language model to predict the type of the first filtered data to obtain the type of the first filtered data, wherein the type is used to characterize whether the first filtered data is related to the vehicle-mounted dialogue system; filtering the conversation data in the first filtered data that is not related to the vehicle-mounted dialogue system based on the type of the first filtered data to obtain the second filtered data.

[0087] The second language model can be used to detect the original dialogue data irrelevant to the vehicle dialogue system to filter out the chat data. The second language model can be one of the following: a recurrent neural network, a long short-term memory network, etc., which is not limited here and can be determined as needed.

[0088] In an optional embodiment, the second language model can be used to filter the dialogue data irrelevant to the vehicle dialogue system, ensuring that the vehicle dialogue system processes user inquiries related to the functions of the vehicle dialogue system, avoiding the processing of information irrelevant to the vehicle dialogue system, thereby improving the performance of the vehicle dialogue system while protecting the privacy of the user, especially when processing dialogue data involving the user's personal privacy, the risk of data leakage can be effectively avoided. Further, the dialogue data irrelevant to the vehicle dialogue system can be filtered to obtain the second filtered data.

[0089] Optionally, based on the initial query data in the second filtered data, target query data is determined, including: vectorizing the initial query data to obtain a query semantic vector; converting the query semantic vector to a three-dimensional data space to obtain a converted semantic vector; performing data aggregation based on the converted semantic vector to obtain at least one cluster center; and determining the initial query data corresponding to at least one cluster center as the target query data.

[0090] In an optional embodiment, the initial query data can be vectorized to obtain a query semantic vector. Optionally, the initial query data can be vectorized by a bag of words model, or the initial query data can be vectorized by a word vector model. Then, the query semantic vector is converted to a three-dimensional data space. By reducing its dimension to three dimensions, a visualization tool in a three-dimensional space, such as a 3D scatter plot, can be used to observe the relationship between the query semantic vectors, helping analysts to more intuitively identify clusters and outliers in the query semantic vector, thereby guiding subsequent data mining and optimization work. Optionally, component analysis can be used to convert the data into a three-dimensional space, or an autoencoder can be used to convert the query semantic vector into a three-dimensional data space to obtain a converted semantic vector. Alternatively, a t-distributed Stochastic Neighbor Embedding (t-SNE) algorithm can be used for three-dimensional space conversion. In the scenario of an in-vehicle dialogue system, the query semantic vector that vectorizes the initial query data often has a high dimension, and it may be difficult to intuitively understand its distribution characteristics by directly analyzing it in a high-dimensional space. Converting the query semantic vector to three-dimensional data space helps to mine user intent and more accurately identify patterns of user needs. It can also reduce the difficulty of subsequent clustering, optimize the performance of clustering algorithms, and improve the accuracy of mining potential user intent in unanswered data. Through vectorization and three-dimensional data space conversion, the in-vehicle dialogue system can more intuitively understand the semantic structure of user queries, thereby more accurately locating problems during the optimization process, especially when dealing with multi-round dialogues and user queries in complex contexts, which can effectively improve the understanding and response capabilities of the in-vehicle dialogue system.

[0091] Optionally, data is aggregated in three-dimensional space based on the converted semantic vectors, and the cluster centers obtained are as follows: Figure 7 As shown, in Figure 7 In the figure, “0”, “1”, “2” and “4” represent four cluster centers respectively.

[0092] by Figure 8 The following describes the process of obtaining target inquiry data when the target type is unanswered data. Figure 8 It is a schematic diagram of an optional method for obtaining target query data according to an embodiment of the present invention.

[0093] like Figure 8As shown, noise detection is performed on the original dialogue data, and the noise data in the original dialogue data is filtered. Then, density clustering is performed to obtain the first filtered data. That is, density clustering is performed on the filtered original dialogue data. Further, the type of the first filtered data is predicted to obtain the second filtered data. Optionally, the type of the first filtered data can be predicted by a second language model to obtain the type of the first filtered data, and the type of the first filtered data can be used to characterize whether the first filtered data is related to the vehicle-mounted dialogue system, and then the dialogue data that is not related to the vehicle-mounted dialogue system is filtered to obtain the second filtered data. Then, data aggregation is performed on the second filtered data to determine the target inquiry data.

[0094] Optionally, the above method also includes: associating the target inquiry data with the knowledge data of the in-vehicle dialogue system to obtain associated knowledge data; optimizing the in-vehicle dialogue system based on the target training data and the associated knowledge data, wherein the target training data is different from the training data of the in-vehicle dialogue system.

[0095] In an optional embodiment, the target query data can be associated with the knowledge data to optimize the in-vehicle dialogue system. This can help the in-vehicle dialogue system to continuously enrich and update the knowledge base, improve the intelligence of the in-vehicle dialogue system, and improve the accuracy and response quality of the in-vehicle dialogue system. Optionally, the target query data can be associated with the knowledge data of the in-vehicle dialogue system by calculating the similarity to obtain the knowledge points related to the target query data in the knowledge data. Further, the knowledge base of the in-vehicle dialogue system can be updated or expanded, for example, new knowledge points can be added, knowledge points can be optimized, or positive and negative examples can be added for each knowledge point. Then, the in-vehicle dialogue system is optimized using target training data that is different from the training data, which can ensure the objectivity and generalization ability of the evaluation results and avoid overfitting the training data.

[0096] Fig. 9 is a schematic diagram of an optional method for processing conversation data according to an embodiment of the present invention, such as Fig. 9 As shown, the original conversation data is obtained from the online log, and further divided into answered data and unanswered data according to the display threshold. The answered data is sent to the front-end for display, including repeated user inquiries and user feedback. The unanswered data is processed, including noise detection, chat filtering and data aggregation, so as to obtain the target inquiry data, and then the target inquiry data is sent to the front-end for display.

[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0098] According to an embodiment of the present invention, a device embodiment of a conversation data processing device is provided. It should be noted that the device can be used to execute the conversation data processing method described above.

[0099] Fig.10 is a schematic diagram of a device for processing conversation data according to an embodiment of the present invention. Fig.10 As shown, the device includes: an acquisition module 1002, used to acquire original dialogue data of the in-vehicle dialogue system, wherein the original dialogue data includes user inquiry data and system response data corresponding to the user inquiry data; a determination module 1004, used to determine the target type of the original dialogue data based on the user inquiry data and the system response data; a mining module 1006, used to perform data mining on the original dialogue data based on the target type to obtain target inquiry data, wherein the target inquiry data is used to optimize the in-vehicle dialogue system.

[0100] Optionally, the determination module is also used to determine the degree of match between the user inquiry data and the system response data; when the degree of match is greater than or equal to a preset threshold, the target type is determined to be answered data; when the degree of match is less than a preset threshold, the target type is determined to be unanswered data.

[0101] Optionally, the mining module is further used to aggregate the original conversation data according to the conversation time of the original conversation data to obtain two adjacent rounds of conversation data; mine the user inquiry data in the previous round of conversation data based on the user feedback data in the latter round of conversation data in the two adjacent rounds of conversation data to obtain first inquiry data; mine the user inquiry data in the previous round of conversation data based on the two user inquiry data in the two adjacent rounds of conversation data to obtain second inquiry data; merge and deduplicate the first inquiry data and the second inquiry data to obtain target inquiry data.

[0102] Optionally, the mining module is also used to input user feedback data and user inquiry data into the prediction model, and use the prediction model to predict whether two adjacent rounds of dialogue data meet user needs to obtain a prediction result; when the prediction result is that two adjacent rounds of dialogue data do not meet user needs, the user inquiry data in the previous round of dialogue data is determined to be the first inquiry data.

[0103] Optionally, the mining module is also used to extract features from the two user query data to obtain two semantic vectors; based on the text similarity between the two user query data and the similarity of the two semantic vectors, determine the target similarity of the two user query data; when the target similarity is greater than the preset similarity, determine that the user query data in the previous round of conversation data is the second query data.

[0104] Optionally, the mining module is also used to determine the text similarity between two user query data to obtain a first similarity result; determine the similarity between two semantic vectors to obtain a second similarity result; and perform weighted processing on the first similarity result and the second similarity result to obtain a target similarity.

[0105] Optionally, the mining module is also used to filter the noise data in the original dialogue data to obtain first filtered data; filter the dialogue data irrelevant to the vehicle dialogue system in the first filtered data to obtain second filtered data; and determine the target inquiry data based on the initial inquiry data in the second filtered data.

[0106] Optionally, the mining module is also used for one of the following: using the first language model to perform noise detection on the original conversation data to obtain the detection result of the original conversation data, and filtering the noise data in the original conversation data based on the detection result to obtain the first filtered data; using a density clustering algorithm to cluster the original conversation data to obtain the clustering result, and deleting other conversation data in the original conversation data except the clustering result to obtain the first filtered data.

[0107] Optionally, the mining module is also used to predict the type of the first filtered data using the second language model to obtain the type of the first filtered data, wherein the type is used to characterize whether the first filtered data is related to the in-vehicle dialogue system; based on the type of the first filtered data, the dialogue data in the first filtered data that is not related to the in-vehicle dialogue system is filtered to obtain the second filtered data.

[0108] Optionally, the mining module is also used to vectorize the initial query data to obtain a query semantic vector; convert the query semantic vector to a three-dimensional data space to obtain a converted semantic vector; perform data aggregation based on the converted semantic vector to obtain at least one cluster center; and determine that the initial query data corresponding to at least one cluster center is the target query data.

[0109] Optionally, the above device also includes an optimization module.

[0110] The optimization module is used to associate the target inquiry data with the knowledge data of the vehicle dialogue system to obtain the associated knowledge data; and optimize the vehicle dialogue system based on the target training data and the associated knowledge data, wherein the target training data is different from the training data of the vehicle dialogue system.

[0111] An embodiment of the present application further provides a vehicle, comprising: a memory storing an executable program; and a processor for running the program, wherein the method in each embodiment of the present invention is executed when the program is running.

[0112] An embodiment of the present application further provides a computer-readable storage medium, which includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0113] An embodiment of the present application further provides a computer program product, including a computer program, which implements the methods in various embodiments of the present invention when executed by a processor.

[0114] An embodiment of the present application further provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0115] The embodiments of the present application further provide a computer program, which implements the methods in the above-mentioned embodiments of the present invention when executed by a processor.

[0116] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0118] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0119] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0120] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0121] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for processing conversation data, characterized in that: include: Acquire original dialogue data of the in-vehicle dialogue system, wherein the original dialogue data includes user inquiry data and system response data corresponding to the user inquiry data; Determining a target type of the original conversation data based on the user query data and the system response data; Data mining is performed on the original dialogue data based on the target type to obtain target query data, wherein the target query data is used to optimize the in-vehicle dialogue system.

2. The method according to claim 1, characterized in that Determining a target type of the original conversation data based on the user query data and the system response data includes: Determining a degree of match between the user query data and the system response data; When the matching degree is greater than or equal to a preset threshold, determining the target type as answered data; When the matching degree is less than the preset threshold, the target type is determined to be unanswered data.

3. The method according to claim 1, characterized in that When the target type is answered data, data mining is performed on the original conversation data to obtain target inquiry data, including: Aggregating the original conversation data according to the conversation time of the original conversation data to obtain two adjacent rounds of conversation data; Based on the user feedback data in the latter round of the two adjacent rounds of dialogue data, mining the user inquiry data in the former round of dialogue data to obtain first inquiry data; Based on the two user inquiry data in the two adjacent rounds of conversation data, mining the user inquiry data in the previous round of conversation data to obtain second inquiry data; The first query data and the second query data are merged and deduplicated to obtain the target query data.

4. The method according to claim 3, characterized in that Based on the user feedback data in the latter round of the two adjacent rounds of dialogue data, the user inquiry data in the former round of dialogue data is mined to obtain the first inquiry data, including: Inputting the user feedback data and the user inquiry data into a prediction model, and using the prediction model to predict whether the two adjacent rounds of conversation data meet the user's needs, to obtain a prediction result; When the prediction result is that the two adjacent rounds of dialogue data do not meet the user's needs, the user inquiry data in the previous round of dialogue data is determined as the first inquiry data.

5. The method according to claim 3, characterized in that: Based on the two user inquiry data in the two adjacent rounds of conversation data, mining the user inquiry data in the previous round of conversation data to obtain second inquiry data includes: Performing feature extraction on the two user query data to obtain two semantic vectors; Determining a target similarity between the two user query data based on the text similarity between the two user query data and the similarity between the two semantic vectors; When the target similarity is greater than a preset similarity, the user inquiry data in the previous round of conversation data is determined to be the second inquiry data.

6. The method according to claim 5, characterized in that Determining target similarity of the two user query data based on text similarity between the two user query data and similarity of the two semantic vectors includes: Determine the text similarity between the two user query data to obtain a first similarity result; Determining the similarity between the two semantic vectors to obtain a second similarity result; The first similarity result and the second similarity result are weighted to obtain the target similarity.

7. The method according to claim 2, characterized in that: When the target type is the unanswered data, data mining is performed on the original conversation data to obtain target inquiry data, including: Filtering the noise data in the original conversation data to obtain first filtered data; filtering the dialogue data irrelevant to the in-vehicle dialogue system in the first filtering data to obtain second filtering data; The target query data is determined based on the initial query data in the second filtered data.

8. The method according to claim 7, characterized in that Filtering the noise data in the original conversation data to obtain first filtered data includes one of the following: Using a first language model to perform noise detection on the original conversation data to obtain a detection result of the original conversation data, and filtering noise data in the original conversation data based on the detection result to obtain the first filtered data; The original conversation data is clustered using a density clustering algorithm to obtain a clustering result, and other conversation data in the original conversation data except the clustering result is deleted to obtain the first filtered data.

9. The method according to claim 7, characterized in that: Filtering the dialogue data irrelevant to the in-vehicle dialogue system in the first filtered data to obtain second filtered data includes: Predicting the type of the first filtered data using the second language model to obtain the type of the first filtered data, wherein the type is used to indicate whether the first filtered data is related to the in-vehicle dialogue system; The dialogue data irrelevant to the in-vehicle dialogue system in the first filtering data is filtered based on the type of the first filtering data to obtain the second filtering data.

10. The method according to claim 7, characterized in that Determining the target query data based on the initial query data in the second filtered data includes: Vectorizing the initial query data to obtain a query semantic vector; Converting the query semantic vector into a three-dimensional data space to obtain a converted semantic vector; Performing data aggregation based on the converted semantic vector to obtain at least one clustering center; Initial query data corresponding to the at least one cluster center is determined as the target query data.

11. The method according to any one of claims 1 to 10, characterized in that The method further comprises: Associating the target query data with the knowledge data of the in-vehicle dialogue system to obtain associated knowledge data; The in-vehicle dialogue system is optimized based on target training data and the associated knowledge data, wherein the target training data is different from the training data of the in-vehicle dialogue system.

12. A device for processing conversation data, characterized in that: include: An acquisition module, used to acquire original dialogue data of the in-vehicle dialogue system, wherein the original dialogue data includes user inquiry data and system response data corresponding to the user inquiry data; A determination module, configured to determine a target type of the original conversation data based on the user query data and the system response data; A mining module is used to perform data mining on the original dialogue data based on the target type to obtain target query data, wherein the target query data is used to optimize the in-vehicle dialogue system.

13. A vehicle, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 11 when running.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 11.

15. A computer program product, characterized in that The invention comprises a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 11.

Images