Data processing method, device and equipment and readable storage medium

Abstract

The invention discloses a data processing method and device, equipment and a readable storage medium. The method comprises the steps of obtaining to-be-recognized text data; performing intention recognition on the text data by defining an intention list and defining a slot position list set to obtain a text intention of the text data; defining slot position lists in the defining slot position list set are in one-to-one correspondence with defining intentions in the defining intention list; performing slot recognition on the text data through a defined slot value list set and a defined slot list corresponding to the text intention to obtain a text slot of the text data; the defined slot position value lists in the defined slot position value list set are in one-to-one correspondence with defined slot positions in the defined slot position lists. The method can be applied to various scenes such as the map field, the traffic field, the automatic driving field, the vehicle-mounted scene, the cloud technology, artificial intelligence, intelligent traffic and auxiliary driving, and the intention recognition accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a data processing method, apparatus, device, and readable storage medium. Background Art

[0002] Natural Language Understanding (NLU) is a general term for all method models or tasks that support machines to understand the content of text. Intent recognition is a technology in natural language understanding, which is used to determine the intent or purpose when a user has a conversation with a computer.

[0003] Currently, the prior art performs natural language understanding on the natural language input by a user to identify the intent expressed by the natural language. For example, a pre-trained neural network model is used to perform semantic parsing on the text input by the user to identify the intent expressed by the semantics.

[0004] However, when the related art performs intent recognition on text data, the reference information used is relatively simple and single. For example, when recognizing an intent, only the defined intents are used as references to screen and analyze the intent of the text data; when recognizing slots, only the defined slots are used as references to screen and analyze the slots that the text data can match; in the process of both intent recognition and slot recognition, the references used are not comprehensive enough, and often the true intent expressed by the natural language cannot be effectively recognized, reducing the overall accuracy of intent recognition. Summary of the Invention

[0005] Embodiments of this application provide a data processing method, apparatus, device, and readable storage medium, which can perform intent recognition by combining information in different dimensions, improving the accuracy of intent recognition.

[0006] An embodiment of this application on the one hand provides a data processing method, including:

[0007] Obtain text data to be recognized;

[0008] Perform intent recognition on the text data through a defined intent list and a set of defined slot lists to obtain the text intent of the text data; there is a one-to-one correspondence between the defined slot lists in the set of defined slot lists and the defined intents in the defined intent list;

[0009] Perform slot recognition on the text data through a set of defined slot value lists and the defined slot list corresponding to the text intent to obtain the text slots of the text data; there is a one-to-one correspondence between the defined slot value lists in the set of defined slot value lists and the defined slots in the defined slot list; the text intent and the text slots are used to reflect the service required by the text data.

[0010] In one aspect of the embodiments of the present application, a data processing device is provided, including:

[0011] A data acquisition module, configured to acquire text data to be recognized;

[0012] An intent recognition module, configured to perform intent recognition on the text data by defining an intent list and a set of defined slot lists to obtain the text intent of the text data; there is a one-to-one correspondence between the defined slot lists in the set of defined slot lists and the defined intents in the defined intent list;

[0013] A slot recognition module, configured to perform slot recognition on the text data by defining a set of slot value lists and the defined slot list corresponding to the text intent to obtain the text slots of the text data; there is a one-to-one correspondence between the defined slot value lists in the set of defined slot value lists and the defined slots in the defined slot list; the text intent and the text slots are used to reflect the services required by the text data.

[0014] In one embodiment, the specific implementation manner of the intent recognition module performing intent recognition on the text data by defining an intent list and a set of defined slot lists to obtain the text intent of the text data includes:

[0015] Performing intent semantic recognition on the text data through the defined intent list to obtain the predicted intent of the text data; the defined intent list includes the predicted intent;

[0016] Obtaining the defined slot list corresponding to the predicted intent from the set of defined slot lists;

[0017] Performing intent adaptability detection on the predicted intent through the defined slot list corresponding to the predicted intent to obtain the adaptability detection result of the predicted intent;

[0018] If it is determined that the adaptability detection result indicates that the predicted intent is adaptable to the text data, then the predicted intent is determined as the text intent of the text data.

[0019] In one embodiment, the specific implementation manner of the intent recognition module performing intent semantic recognition on the text data through the defined intent list to obtain the predicted intent of the text data includes:

[0020] Obtaining historical text data having a context coherence association relationship with the text data;

[0021] Generating intent recognition prompt information according to the defined intent list, the historical text data and the text data, and inputting the intent recognition prompt information into a language understanding model;

[0022] In the language understanding model, performing intent semantic recognition on the text data based on the intent recognition prompt information, and recognizing the predicted intent of the text data from the defined intent list.

[0023] In one embodiment, the intention recognition module performs intention adaptability detection on the predicted intention by predicting the defined slot list corresponding to the intention, and the specific implementation method for obtaining the adaptability detection result of the predicted intention includes:

[0024] Determine the defined slot list corresponding to the predicted intention as the first target slot list;

[0025] Obtain the intention description information configured for the predicted intention and the historical text data having a contextually coherent association with the text data;

[0026] Generate intention detection prompt information based on the predicted intention, the intention description information of the predicted intention, the first target slot list, the historical text data, and the text data, and input the intention detection prompt information into the language understanding model;

[0027] In the language understanding model, analyze the adaptability between the text data and the predicted intention based on the intention detection prompt information to obtain the adaptability detection result of the predicted intention.

[0028] In one embodiment, the slot recognition module performs slot recognition on the text data by defining a set of slot value lists and the defined slot list corresponding to the text intention, and the specific implementation method for obtaining the text slot of the text data includes:

[0029] Determine the defined slot list corresponding to the text intention as the second target slot list;

[0030] Perform slot matching recognition on the text data through the second target slot list to obtain the predicted slot of the text data; the second target slot list includes the predicted slot;

[0031] Obtain the defined slot value list corresponding to the predicted slot from the set of defined slot value lists;

[0032] Perform matching detection on the predicted slot through the defined slot value list corresponding to the predicted slot to obtain the matching detection result of the predicted slot;

[0033] If it is determined that the matching detection result indicates that the predicted slot matches the text data, then determine the predicted slot as the text slot of the text data.

[0034] In one embodiment, the specific implementation method for the slot recognition module to perform slot matching recognition on the text data through the second target slot list to obtain the predicted slot of the text data includes:

[0035] Obtain the intention description information configured for the text intention and the historical text data having a contextually coherent association with the text data;

[0036] Generate slot matching prompt information based on the text intention, the intention description information of the text intention, the second target slot list, the historical text data, and the text data, and input the slot matching prompt information into the language understanding model;

[0037] In the language understanding model, perform slot matching recognition on the text data based on the slot matching prompt information to obtain a slot recognition result; the slot recognition result includes the predicted slot matched and recognized by the language understanding model from the second slot list, the text slot value in the text data that matches the predicted slot, and the reasoning description corresponding to the predicted slot

[0038] Determine the predicted slot included in the slot recognition result as the predicted slot of the text data.

[0039] In one embodiment, the specific implementation manner of the slot recognition module to perform a matching detection on the predicted slot through the defined slot value list corresponding to the predicted slot to obtain the matching detection result of the predicted slot includes:

[0040] Obtain historical text data that has a context coherent association relationship with the text data;

[0041] Generate slot detection prompt information based on the text intention, the predicted slot under the text intention, the defined slot value list corresponding to the predicted slot, the text slot value in the text data that matches the predicted slot, the reasoning description corresponding to the predicted slot, the historical text data, and the text data, and input the slot detection prompt information into the language understanding model;

[0042] In the language understanding model, detect the matching between the text data and the predicted slot based on the slot detection prompt information to obtain the matching detection result of the predicted slot.

[0043] In one embodiment, the specific implementation manner of the slot recognition module to detect the matching between the text data and the predicted slot in the language understanding model based on the slot detection prompt information to obtain the matching detection result of the predicted slot includes:

[0044] In the language understanding model, determine the text slot value in the text data that matches the predicted slot as the verification slot value;

[0045] Determine the defined slot value list corresponding to the predicted slot as the target slot value list;

[0046] Perform a similarity analysis on the verification slot value and the target slot value list to obtain the similarity analysis result between the verification slot value and the target slot value list;

[0047] If the similarity analysis result indicates that there is a similarity between the verification slot value and the target slot value list, determine that the matching detection result of the predicted slot is a passed matching detection result;

[0048] If the similarity analysis result indicates that there is no similarity between the verification slot value and the target slot value list, it is determined that the matching detection result of the predicted slot is a non-passing result of the matching detection.

[0049] In one embodiment, the specific implementation manner for the slot recognition module to perform a similarity analysis on the verification slot value and the target slot value list to obtain the similarity analysis result between the verification slot value and the target slot value list includes:

[0050] Calculate the word similarity between the verification slot value and each defined slot value in the target slot value list to obtain a set of word similarities;

[0051] Traverse the set of word similarities;

[0052] If there is a word similarity greater than the similarity threshold in the set of word similarities, it is determined that there is a defined slot value similar to the verification slot value in the target slot value list, and the similarity analysis result between the verification slot value and the target slot value list is determined as a result of the existence of similarity;

[0053] If there is no word similarity greater than the similarity threshold in the set of word similarities, it is determined that there is no defined slot value similar to the verification slot value in the target slot value list, and the similarity analysis result between the verification slot value and the target slot value list is determined as a result of the non-existence of similarity.

[0054] In one embodiment, the target slot value list contains the defined slot value S i , and the set of word similarities contains the word similarity between the verification slot value and the defined slot value S i ; i is a positive integer;

[0055] The specific implementation manner for the slot recognition module to calculate the word similarity between the verification slot value and each defined slot value in the target slot value list to obtain a set of word similarities includes:

[0056] Obtain a first word vector for characterizing the verification slot value and a second word vector for characterizing the defined slot value S i ;

[0057] Calculate the vector distance between the first word vector and the second word vector;

[0058] Determine the word similarity between the verification slot value and the defined slot value S i according to the vector distance.

[0059] In one embodiment, the specific implementation manner for the slot recognition module to determine the word similarity between the verification slot value and the defined slot value S i according to the vector distance includes:

[0060] Obtain a distance mapping table; the distance mapping table contains the mapping relationship between the configured distance interval set and the similarity set, and there is a mapping relationship between a configured distance interval in the configured distance set and a similarity in the similarity set;

[0061] Determine the configured distance interval to which the vector distance in the configured distance interval set belongs as the target configured distance interval;

[0062] Determine the similarity in the similarity set that has a mapping relationship with the target configured distance interval as the word similarity between the verification slot value and the defined slot value S i between.

[0063] In one embodiment, after obtaining the text slots of the text data, the data processing device further includes:

[0064] An intent determination module, configured to determine the true intent indicated by the text data according to the text intent and the text slots;

[0065] A service module, configured to provide services to the text providing object according to the true intent indicated by the text data; the text providing object refers to the object that provides the text data.

[0066] In one embodiment, the true intent is a navigation intent; the navigation intent refers to the intent to obtain a navigation path from a starting position point to a destination position point;

[0067] The specific implementation manner in which the service module provides services to the text providing object according to the true intent indicated by the text data includes:

[0068] Generate a set of navigation paths from the starting position point to the destination position point;

[0069] Perform a consumption duration analysis on each navigation path in the navigation path set to obtain the path consumption duration required for each navigation path;

[0070] Generate service indication information including the navigation path set and the path consumption duration required for each navigation path;

[0071] Push the service indication information to the text providing object.

[0072] One aspect of the embodiments of the present application provides a computer device, including: a processor and a memory;

[0073] The memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method in the embodiments of the present application.

[0074] In one aspect of the embodiments of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, the methods in the embodiments of the present application are executed.

[0075] In one aspect of the present application, a computer program product is provided. The computer program product includes a computer program, and the computer program is stored in a computer-readable storage medium. The processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method provided in one aspect of the embodiments of the present application.

[0076] In the embodiments of the present application, after obtaining the text data to be recognized, on the one hand, the defined intent list and the set of defined slot lists can be combined to jointly perform intent recognition on the text data to obtain the text intent of the text data. That is to say, in the recognition service of the dimension of recognizing the text intent of the text data, not only the information of the defined intent dimension is used, but also the information of the defined slot dimension corresponding to the defined intent is combined. The defined slot list set is used to jointly perform intent recognition and analysis on the text data. Compared with the text intent obtained only by using the defined intent list, the text intent analyzed by combining the defined slot list set can be more accurate. On the other hand, after recognizing the text intent of the text data, it is supported to combine the defined slot value list set and the defined slot list corresponding to the text intent to jointly perform slot recognition on the text data to obtain the text slot of the text data. Then, in the recognition service of the dimension of recognizing the text slot of the text data, not only the information of the defined slot dimension is used, but also the information of the defined slot value dimension corresponding to the defined slot is combined. The multi-dimensional information is used to jointly perform slot analysis on the text data. Compared with the text slot obtained only by using the defined slot list, the text slot obtained by combining the defined slot value list set can be more accurate. It can be seen that the embodiments of the present application introduce defined slots in the text intent recognition of text data, and the text intent of the text data can be more accurately analyzed and recognized through the defined slots corresponding to the defined intent; defined slot values are introduced in the text slot recognition of text data, and the text slot of the text data can be more accurately analyzed and recognized through the defined slot values corresponding to the defined slots. When the text intent and text slot of the text data are both highly accurate, the true intent determined according to the text intent and text slot can also be more accurate. In summary, when the present application performs intent recognition, for content recognition in any dimension (such as text intent recognition, text slot recognition), the existing information of other relevant dimensions is combined as a reference, which can enrich the reference information in the recognition process, improve the recognition effect of natural language understanding, and improve the intent recognition ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0078] Figure 1 It is a schematic structural diagram of an intention recognition system provided by an exemplary embodiment of the present application;

[0079] Figure 2 It is a schematic flowchart of a data processing method provided by an exemplary embodiment of the present application;

[0080] Figure 3 It is a schematic diagram of a brief scenario for text intention recognition provided by an embodiment of the present application;

[0081] Figure 4 It is a schematic diagram of a brief scenario for text slot recognition provided by an embodiment of the present application;

[0082] Figure 5 It is a schematic flowchart of a process for text intention recognition provided by an embodiment of the present application;

[0083] Figure 6 It is a schematic flowchart of a process for text slot recognition provided by an embodiment of the present application;

[0084] Figure 7 It is a schematic logical structure diagram provided by an embodiment of the present application;

[0085] Figure 8 It is a schematic structural diagram of a data processing device provided by an embodiment of the present application;

[0086] Figure 9 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0087] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0088] The embodiments of the present application relate to artificial intelligence and related technologies. For ease of understanding, the following will first briefly elaborate on artificial intelligence and related technical terms and concepts.

[0089] 1. Artificial Intelligence (AI)

[0090] Artificial intelligence uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, including theories, methods, technologies, and application systems for perceiving the environment, acquiring knowledge, and using knowledge to achieve the best results. Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0091] Furthermore, the embodiments of this application mainly involve technologies such as Machine Learning (ML) and Natural Language Understanding (NLU) in artificial intelligence technology. Among them: Machine learning is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. Machine learning specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning. Natural language understanding, commonly known as artificial conversation, uses electronic computers to simulate the language communication process of humans, enabling computers to understand and use natural languages of human society (such as Chinese or English, etc.) to achieve natural language communication between humans and computers; to replace some of the mental labor of humans, including querying information, answering questions, extracting literature, compiling materials, and all processing of natural language information.

[0092] 2. Intent Recognition

[0093] Intention recognition, also known as intention classification, aims to use technologies such as machine learning and natural language understanding in the field of artificial intelligence mentioned above to associate the text in the form of natural language input by the user with the given intention classification, so as to identify the user's intention and provide the user with better intelligent question-answering capabilities. In short, intention recognition, as the name implies, is the process of judging the true intention of the user by performing semantic analysis (or called semantic recognition) on the natural language (such as Chinese or English, etc.) input by the user. For example, in the intelligent question-answering scenario, when the text in the form of natural language input by the user "Query the weather in City A today" is obtained, semantic recognition can be performed on this text to identify the true intention expressed by this text, and then provide the user with corresponding services based on this true intention (such as providing the user with the temperature in City A today, whether it is sunny or rainy, etc.).

[0094] In practical applications, the process of performing intent recognition on the text input by the user generally includes intent classification and slot filling. Among them: ① Intent classification is mainly based on multiple pre-configured optional intent categories (the optional intent categories can be defined and configured manually in advance. The optional intent categories can be called defined intents. Defined intents such as weather query, restaurant query, tourist attraction query, computer query, tutoring class query, hospital query, academic question answer query, movie query, etc.). The intent expressed by the text input by the user is classified to determine the intent category to which the text belongs. ② For a computer device, after obtaining the intent category to which the intent expressed by the text belongs, it needs to obtain one or more intent slots corresponding to the intent category (or called attributes. In a task-based dialogue, it is usually necessary to collect the necessary user information through multiple rounds of interaction to complete the dialogue. Slots are the abstraction of intents for user information. For example, in the intent category of [weather query], it is necessary to know the location and time that the user wants to query. Therefore, two slots of time and location are required. Intent slots can be defined and configured manually. In this application, the defined intent slots can be called defined slots), and determine the slot values (or called attribute values) of the slots corresponding to the intent category from the text; in this way, the computer device can fill the slot values into the corresponding slots, so that the computer device can execute the user's true intent based on the slot values (for example, for the intent category of [weather query], its defined slots can include a time slot and a location slot. Then, in the above example text "Query the weather in City A today", "today" can be the slot value of the time slot, and "City A" can be the slot value of the location slot. Based on the time slot and the location slot, the true intent of the user can be determined as: need to query the weather in City A today). For example, assume that the text input by the user is "Book a movie ticket for the movie MT starring LH". After semantic analysis of this text, the intent expressed by this text can be classified into the intent category of [book movie ticket]; further, the intent slots corresponding to the intent category of [book movie ticket] can include but are not limited to: <actor>, <movie name>, and <cinema>, etc.; among them, the slot values of the intent slots <actor> and <movie name> can be recognized from the text. For example, the slot value of the intent slot <actor> is "LH", and the slot value of the intent slot <movie name> is "MT". However, the slot value of the intent slot <cinema> cannot be recognized from the text, so it is necessary to ask the user through the computer device, so as to further determine that the slot value of the intent slot <cinema> is "C" according to the newly input text "Cinema C" by the user. It should be noted that when configuring and defining intents and intent slots in this application, one or more intents can be configured and defined, and one or more intent slots can also be configured and defined for each intent. Then, the number of defined intents obtained is one or more, and each defined intent can correspond to one or more defined slots.During the intent recognition process, only the pre-set defined intents need to be called for intent recognition and the defined slots under the corresponding intent classification need to be called for slot filling.

[0095] Based on the above, during the process of intent recognition for the text input by the user, it is necessary to first classify the text based on the defined intents to identify the intent expressed by the text (the intent expressed by the text can be understood as a relatively broad field or direction reflected by the text. For example, in the weather query field, restaurant query field, tourist attraction query field, computer query field, tutoring class query field, hospital query field, academic question query field, movie query field, flight reservation field, hotel reservation field, etc. In this application, the intent expressed by the text can be referred to as the text intent); further, after identifying the text intent of the text, it is necessary to fill the slots based on the defined slots corresponding to the text intent to identify the user's true intent. In the process of traditional intent recognition technology for identifying the text intent of the text, only the information of one dimension of the defined intent is used for intent classification. For example, only the defined intent is used as a reference to identify and analyze the text data to screen out the text intent of the text data from each defined intent; when filling the slots, only the information of one dimension of the defined slots is used for slot matching and filling. For example, only the defined slots are used as a reference to match and analyze the text data to screen out the slots that the text data can match as the text slots. It can be seen that whether it is intent classification or slot matching, the information referred to is relatively simple and single, not comprehensive enough, and the accuracy of the identified text intent and slot filling results is not high enough, resulting in a decrease in the accuracy of intent recognition. In addition, the traditional intent recognition technology uses a pre-trained neural network model for intent recognition. In order to improve the accuracy of the model output results, a large amount of dialogue data with coherent context will be configured to fine-tune the model (such as using machine learning methods to train and learn the model to fine-tune the model parameters). This method will take a lot of time and effort to configure the dialogue data, and the data configuration threshold is relatively high; at the same time, if the defined intent is updated (such as added), for the model, it will take time to re-learn and fine-tune the model so that the model can accurately perform intent recognition based on the updated defined intent. It can be seen that once the defined intent is updated, the historically trained model cannot be applied to intent recognition in real time, and it needs to be re-learned and fine-tuned, resulting in a decrease in intent recognition efficiency.

[0096] To improve the accuracy of intent recognition and, at the same time, enhance the real-time performance of intent recognition (equivalent to improving the efficiency of intent recognition), this application provides a text intent recognition solution that can combine information from different dimensions for intent recognition (including recognizing the text intent and text slots of the text to determine the true intent of the text), improving the accuracy of intent recognition. This solution does not require training a model. After configuring the defined intents, the defined slots corresponding to each defined intent, and the defined slot values corresponding to each defined slot, it can be applied to the intent recognition business, improving the efficiency of intent recognition. Among them, the intent recognition of the text involved in this solution can include three consecutive steps: 1. Perform intent recognition on the text to obtain the text intent of the text (the text intent is equivalent to identifying the field or direction to which the user's needs belong); 2. Based on the text intent, perform slot recognition on the text to obtain the text slots of the text; 3. Combine the text intent and the text slots to determine the true intent expressed by the text (that is, the specific needs in the identified field or direction by the user).

[0097] Specifically, the intent recognition solution for text provided in the embodiments of this application generally includes the following steps: First, obtain the text data to be recognized input by the user. After obtaining the text data to be recognized, it is possible to combine a defined intent list (that is, a list composed of pre-configured defined intents, which may include one or more defined intents) with a set of defined slot list collections (the set of defined slot list collections may include one or more defined slot lists, where one defined slot list is configured for a certain defined intent, and the defined slot list contains one or more defined slots configured for this defined intent) to jointly perform intent recognition on the text data to obtain the text intent of the text data. After recognizing the text intent of the text data, it is possible to obtain the defined slot list corresponding to the text intent. By combining the defined slot list corresponding to the text intent with the pre-configured set of defined slot value lists (the set of defined slot value lists may include one or more defined slot value lists, where one defined slot value list is configured for a certain defined slot, and the defined slot value list contains one or more defined slot values configured for this defined slot. For example, for the defined slot <Most suitable group>, the configured slot values include "Couple dating", "Friends' outings", "Family parent-child", "Elders' companionship", then each of the above slot values can be used as the defined slot value of the defined slot <Most suitable group>, and these defined slot values can form a defined slot value list {"Couple dating", "Friends' outings", "Family parent-child", "Elders' companionship"}), perform slot recognition on the text data to obtain the text slot of the text data. Further, based on the text intent and the text slot, the true intent that the user input text data wants to express can be accurately analyzed. This true intent can directly and effectively reflect the user's needs (such as what kind of service the user needs). Then, based on this true intent, the computer device can provide effective services that meet the user's needs.

[0098] It can be seen that in the process of intent recognition of text in the embodiments of this application, when recognizing the text intent of the text data, the defined slot list corresponding to each defined intent is introduced, and when recognizing the text slot of the text data, the defined slot value list corresponding to each defined slot is introduced. In the recognition process of each dimension, relevant information of other dimensions closely related to the current dimension is introduced. Thus, the recognition accuracy of the text intent and the text slot can be well improved. In addition, by the method of combining relevant information of other dimensions for recognition, there is no need to train a model. Even if any defined information (such as defined intent, defined slot, defined slot value) is updated, real-time intent recognition can be performed based on the updated information, which can improve the efficiency of intent recognition.

[0099] The intent recognition solution provided by the embodiments of the present application can be applied to any application scenario that requires intent recognition, including but not limited to: intelligent dialogue scenarios, search scenarios, and so on. Among them:

[0100] An intelligent dialogue scenario can refer to a scenario where a person and a computer device communicate with each other using methods such as voice or text; including but not limited to: dialogue scenarios in the fields of intelligent transportation, intelligent vehicles (such as in-vehicle intelligent assistants), and intelligent robots (such as physical robots, or robots in conversation applications (text robots, voice robots, multi-modal digital humans, intelligent quality inspection, agent assistance, etc.)). For example, the dialogue scenario where an intelligent robot in a hotel (or other service scenarios such as customer service) communicates with a human; or, the dialogue scenario where an in-vehicle application communicates with a human; and so on. It should be noted that in an intelligent dialogue scenario, the dialogue between a person and a computer device (such as an intelligent robot with dialogue capabilities) can be a single dialogue or multiple dialogues, and the embodiments of the present application do not limit this. In an intelligent dialogue scenario, the text data to be recognized in the embodiments of the present application is the various text data input by the user during the dialogue.

[0101] A search scenario can refer to a process where a user inputs a search text, and a computer device performs semantic recognition on the search text to provide a search result for the user based on the semantic recognition result of the search text; including but not limited to: various search fields such as the commodity trading field, the advertising search field, and the video search field. Taking the video search field as an example, the user may input a search text with a negative semantics (such as searching for movies not starring A). At this time, the intent recognition solution provided by the embodiments of the present application can accurately recognize the negative intent of the search text, so as to filter out the movies starring actor A from the video database (such as a database for storing videos) and push them to the user. In a search scenario, the text data to be recognized in the embodiments of the present application is the search text input by the user during the search process.

[0102] In summary, the intent recognition solution for text provided by the embodiments of the present application can perform intent recognition by combining information from different dimensions, has high recognition accuracy and real-time performance, and effectively improves the business coverage to a certain extent (such as expanding the applicable scenarios).

[0103] It should be noted that the above-mentioned application scenarios are only examples and do not limit the application scenarios applicable to the intent recognition solution for text provided by the embodiments of the present application.

[0104] Furthermore, the intent recognition solution provided by the embodiments of the present application can be executed by a computer device, which can include a terminal or a server, or both a terminal and a server. To facilitate understanding of the intent recognition solution provided by the embodiments of the present application, the following combinesFigure 1 The intention recognition system shown introduces the application scenarios related to the embodiments of the present application; among them, Figure 1 is a schematic architecture diagram of an intention recognition system provided by an exemplary embodiment of the present application. As Figure 1 shown, the intention recognition system includes a terminal 101 and a server 102; where:

[0105] 1) The terminal 101 may include the terminal device used by the user. Of course, according to the different application scenarios and fields to which the intention recognition solution is applied, the terminals providing the text recognition solution provided by the embodiments of the present application are different. The terminal device may include but is not limited to: smart phones (such as smart phones deployed with the Android system or smart phones deployed with the Internetworking Operating System (IOS)), tablet computers, portable personal computers, Mobile Internet Devices (MID), vehicle-mounted devices, head-mounted devices, smart home and smart voice interaction devices, etc. The embodiments of the present application do not limit the type of the terminal device and are hereby explained.

[0106] For example: in the intelligent robot scenario, the terminal device may be an intelligent robot; that is to say, in this implementation manner, the intention recognition solution provided by the embodiments of the present application can be deployed on the intelligent robot; when the user converses with the intelligent robot, the intelligent robot performs intention recognition on the text input by the user, and after correctly recognizing the true intention of the user, provides services for the user according to the true intention of the user (such as the intelligent robot in a hotel provides services such as guiding the way or fetching meals). Another example: in the intelligent vehicle-mounted scenario, the application program deployed with the intention recognition solution provided by the embodiments of the present application is a vehicle-mounted application program; the types of the vehicle-mounted application program may include but are not limited to: music, video or games, etc.

[0107] Among them, an application program may refer to a computer program for completing a certain or multiple specific tasks; classifying application programs according to different dimensions (such as the running mode, function, etc. of the application program), the types of the same application program under different dimensions can be obtained. For example: classified according to the running mode of the application program, the application program may include but is not limited to: the client installed in the terminal, the applet that can be used without downloading and installation (as a subroutine of the client), the World Wide Web (Web) application program opened through the browser, and so on. Another example: classified according to the function type of the application program, the application program may include but is not limited to: the Instant Messaging (IM) application program, the content interaction application program, the audio application program, or the video application program, and so on. Among them, the instant messaging application program refers to an application program for instant communication messages and social interaction based on the Internet. The instant messaging application program may include but is not limited to: the application program with communication functions, the map application program with interaction functions, the game application program, and so on. The content interaction application program refers to an application program that can realize content interaction, such as the sharing platform, personal space, news and other application programs. The audio application program refers to an application program that realizes audio functions based on the Internet. The audio application program may include but is not limited to: the music application program with music playing and editing capabilities, the radio application program with radio playing capabilities, or the live broadcast application program with live broadcast capabilities, and so on. The video application program refers to an application program that can play pictures. The video application program may include but is not limited to: the application program with short videos (the video length is often short, such as a few seconds or a few minutes, etc.), the application program with long videos (such as videos with a long playing time similar to movies or TV dramas), and so on.

[0108] Of course, the intention recognition solution provided by the embodiments of the present application can be directly deployed on a device (such as a smart robot) or outside the application program as described above, and can also be deployed in the device or application program in the form of a plugin. The embodiments of the present application do not limit the carrier for deploying the text recognition solution.

[0109] 2) The server 102 can be the server corresponding to the terminal, used to interact with the terminal to provide computing and application service support for the terminal. Specifically, this server is the background server corresponding to the application deployed in the terminal, used to interact with the terminal to provide computing and application servers for the application. Among them, the server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0110] Among them, the terminal 101 and the server 102 can be directly or indirectly connected through wired or wireless communication methods, which are not limited in this application. In addition, the number of terminals and servers is not limited in the embodiments of this application; in Figure 1 it is only an example that the number of both the terminal 101 and the server 102 is single, and in actual applications, it may include multiple servers with distributed distribution, which is specifically stated here.

[0111] The following introduces the general process of the intent recognition solution in the application scenario in combination with Figure 1 the shown intent recognition system. In specific implementation, first, if a user wants the terminal (specifically, the application deployed in the terminal, such as a music type application) to execute a certain intent, then the user can input the corresponding text data to the terminal. After the terminal obtains the text data to be recognized, it can transmit the text to the server. After the server receives the text data, it can obtain the defined intent list, the defined slot list corresponding to each defined intent, and the defined slot value list corresponding to each defined slot. By combining the defined intent list with the set of defined slot lists, the text data can be intent-recognized to obtain the text intent of the text data; by combining the defined slot list corresponding to the text intent and the set of defined slot value lists, the text data can be slot-recognized to obtain the text slot of the text data; according to the text intent and the text slot, the server can determine the true intent of the text data. Then, this true intent can reflect the true needs of the user (which can present the service required by the user), so the server can provide a service that meets the true needs of the user based on this true intent. For example, if the true intent of the user is to search for the songs of a certain singer, then the service required by the user is to obtain the songs of a certain singer. Then the server can obtain all the songs of this singer and return the obtained songs to the terminal, and the terminal can display all the songs of this singer to execute or realize the true intent of the user.

[0112] Based on the intention recognition solution and system architecture described above, the following points need to be explained:

[0113] ① As mentioned above in the embodiments of this application Figure 1 The system shown is to more clearly illustrate the technical solution of the embodiments of this application and does not constitute a limitation on the technical solution provided by the embodiments of this application. Those of ordinary skill in the art know that with the evolution of the system architecture and the emergence of new business scenarios, the technical solution provided by the embodiments of this application is equally applicable to similar technical problems. For example, the above takes the execution entity of the embodiments of this application, "computer device", including a terminal and a server as an example, that is, taking the text recognition solution provided by the embodiments of this application jointly executed by the terminal and the server as an example, to introduce an application scenario of the text recognition solution; it should be understood that in actual applications, the computer device can also be a terminal or a server, that is to say, it supports the intention recognition solution provided by the embodiments of this application to be executed by the terminal or the server alone.

[0114] ② The embodiments of this application support using a model with natural language understanding capabilities (such as an LLM model) to implement the intention recognition solution described above. Specifically, this application can use the defined intention list, the defined slot list corresponding to each defined intention, and the defined slot value list corresponding to each defined slot as model parameters and deploy them in the model, and this model does not need to be trained and can be directly deployed in a computer device; in this way, when the computer device needs to perform intention recognition on the text data to be recognized (such as recognizing text intention, recognizing text slots), it can directly call this model to execute the intention recognition solution. Among them, if the computer device used to execute the intention recognition solution provided by the embodiments of this application is a terminal, then this model can be deployed in this terminal. If the computer device used to execute the intention recognition solution provided by the embodiments of this application is a server, then this model is deployed in this server; in this case, the terminal used by the user transmits the text data to be recognized to the server for intention recognition processing, and the server pushes the recognition result to the terminal to provide corresponding services for the user.

[0115] ③ In the embodiments of this application, the collection and processing of relevant data should be strictly in accordance with the requirements of relevant laws and regulations. Obtaining personal information requires the informed consent of the personal subject (or having a legal basis for information acquisition), and subsequent data use and processing behaviors should be carried out within the scope authorized by laws and regulations and the personal information subject. For example, when the embodiments of this application are applied to specific products or technologies, such as obtaining the text data of users, the permission or consent of the users needs to be obtained, and the collection, use, and processing of relevant data (such as the collection and release of bullet screens published by the object, etc.) need to comply with the relevant laws, regulations, and standards of the relevant region.

[0116] Based on the intention recognition solution described above, embodiments of the present application propose a more detailed intention recognition method. The following will introduce the intention recognition method proposed by the embodiments of the present application in detail with reference to the accompanying drawings.

[0117] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of a data processing method provided by an exemplary embodiment of the present application. This process may refer to the process of the intention recognition method provided by the embodiments of the present application. This data processing method (intention recognition method) can be executed by a computer device in the aforementioned system. For example, the computer device is a terminal and / or a server. This data processing method may at least include the following steps S101 - S103:

[0118] Step S101, obtain the text data to be recognized.

[0119] In the present application, the text data to be recognized may be text content generated by a user and carrying the user's intention. Specifically, the acquisition methods of the text data to be recognized may include but are not limited to: ① direct input by the user; for example, the computer device is equipped with a display screen, so that the user can directly input text on the display screen through the physical keyboard (such as an external keyboard) or virtual keyboard of the computer device. ② obtained by converting the voice output by the user; for example, the user can emit a voice signal, so that the computer device can collect the voice signal in the physical environment where the user is located through a microphone and perform text conversion on the voice signal to obtain the text corresponding to the voice output by the user.

[0120] Among them, the text data to be recognized includes one or more characters. The characters may include at least one of the following: Chinese characters (i.e., Chinese words), English characters (i.e., letters), numbers, and punctuation marks (such as commas ",", periods ".", square brackets "【】"). In the actual intention recognition process of the embodiments of the present application, the types and quantities of the characters included in the text data to be recognized obtained are not limited.

[0121] Step S102, perform intention recognition on the text data by defining an intention list and a set of defined slot lists to obtain the text intention of the text data; there is a one-to-one correspondence between the defined slot lists in the set of defined slot lists and the defined intentions in the defined intention list.

[0122] In this application, a defined intent may refer to an intent (which can be understood as a field or direction) pre-configured and defined by a user (specifically, the designed user of the intent recognition system). For example, defined intents may include, but are not limited to: tourist attractions, means of transportation (such as airplanes, cars, trains), electronic items (such as computers, mobile phones, tablets), educational content (such as tutoring classes, tutoring books), medical content (such as hospitals, rescue measures, etc.). Here, the defined intents can be specified and defined based on actual business needs. This application does not specifically limit the defined intents and will not give further examples one by one. The list of defined intents in this application may refer to the list composed of all defined intents.

[0123] A defined slot may refer to a slot configured and defined for a defined intent. Among them, a slot is an abstraction of text data by an intent and can reflect a certain attribute under the defined intent. For example, under the intent (this field) of [weather], if the user's true intent is to query the weather, then it is necessary to know the location and time that the user wants to query in order to query the specific weather. Therefore, for the intent of [weather], both time and location can be used as slots to reflect two different attribute contents under [weather]. This application can allocate and configure different slots for each defined intent. Then, one defined intent can correspond to one or more defined slots. The one or more defined slots corresponding to one defined intent can form a defined slot list, and all the defined slot lists can be combined into a defined slot list set. In short, a defined slot list in the defined slot list set in this application is configured and defined for one defined intent and contains one or more defined slots configured and defined for this defined intent. For example, for the defined intent of [weather], <time> and <location> can be configured and defined as its defined slots; for the defined intent of [tourist attractions], <area>, <type of scenic spot>, <most suitable group of people>, <consumption>, <whether directly accessible by subway>, <ticket price>, <rating>, <opening hours> can be configured and defined as its defined slots; for the defined intent of [restaurant], <area>, <cuisine>, <price>, <consumption>, <whether directly accessible by subway>, <per capita consumption>, <rating>, <business hours> can be configured and defined as its defined slots. For the defined slots of each defined intent, they can be configured and defined based on specific business needs and will not be further exemplified one by one here.

[0124] In the specific implementation of this application, when identifying the text intent of text data, not only the defined intent list is used, but also the defined slot list set is combined to jointly identify the text intent, so as to accurately identify the field or direction to which the text data belongs. The specific implementation of combining the defined slot list set to identify the text intent can at least include two parts: 1) First, perform intent semantic recognition on the text data through the defined intent list, so that the intent that adapts to the text data can be filtered out in the defined intent list (that is, the predicted possible intent, which can be called the predicted intent). Among them, this step can be implemented by a model with natural language understanding ability (this model can be called the language understanding model), and the language understanding model can include but is not limited to: a large language model (Large Language Model, abbreviated as LLM) with emergent capabilities (context understanding, instruction following, step-by-step reasoning), and this large language model has the characteristic of large model parameters and is designed to understand and generate human language; 2) Obtain the defined slot list corresponding to the predicted intent from the defined slot list set, and then detect whether the predicted intent filtered out in the previous step meets the requirements (that is, whether it adapts to the text data) through this defined slot list. In other words, the defined slot list can be used to perform adaptability detection on the predicted intent filtered out in the previous step to filter out the unadapted (that is, non-conforming) predicted intent. The remaining predicted intent after the final filtering is the final text intent of the text data. Among them, this step can be implemented by the above-mentioned language understanding model.

[0125] Please refer to Figure 3 , Figure 3 which is a brief scenario schematic diagram of text intent recognition provided by an embodiment of this application. As Figure 3 shown, assume that in an intelligent dialogue scenario, the user inputs the text data "A friend comes to City A to find me, and I'm going to take her to visit Ancient Town B", and this text data can be used as the text data to be recognized. For this text data to be recognized, the defined intent list can be obtained through language understanding ability (here assume it includes [Tourist Attractions], [Restaurants], [Hotels], [Computers], [Trains], [Weather], [Planes]), and the language understanding model can, based on its own capabilities, combine historical text data to understand the text data to be recognized, so as to analyze the predicted intent of the current text data from the defined intent list. As Figure 3 shown, assume that the possible intents (that is, the predicted intents) recognized and predicted by the model include [Tourist Attractions] and [Restaurants]. Among them, the reason for the model's prediction of [Tourist Attractions] is that "the user mentioned taking a friend to visit Ancient Town B, and the user's intent is to find tourist attractions"; the reason for the prediction of [Hotels] is that "the user mentioned that a friend comes to City A to find him, and accommodation may be needed, and the intent is to find a hotel".

[0126] Further, for the two predicted intents screened out above, an adaptability detection can be performed by defining a set of slot list (the steps with accurate predicted intents output by the model can be filtered and deleted through the adaptability detection). Specifically, the model can obtain the defined slot list corresponding to [tourist attractions] from the defined set of slot lists, and then perform an adaptability detection on the predicted intent [tourist attractions] through the defined slot list; it can also obtain the defined slot list corresponding to [hotel] from the defined set of slot lists, and then perform an adaptability detection on the predicted intent [hotel] through the defined slot list. It should be noted that only the predicted intent that passes the adaptability detection (that is, the adaptability detection result is the result of passing the adaptability detection) can be used as a text intent of the text data; while the predicted intent that fails to pass the adaptability detection (that is, the adaptability detection result is the result of failing to pass the adaptability detection) will be filtered and deleted. Identifying the predicted intent through the defined intent list is equivalent to an initial screening process of the intent. Performing an adaptability detection on the predicted intent through the defined slot list corresponding to the predicted intent is equivalent to a process of fine-filtering the predicted intent screened out above. Through the initial screening and fine-filtering, the recognition accuracy of the text intent of the text data can be improved.

[0127] Step S103, perform slot recognition on the text data through the defined set of slot value lists and the defined slot list corresponding to the text intent to obtain the text slot of the text data; there is a one-to-one correspondence between the defined slot value list in the defined set of slot value lists and the defined slot in the defined slot list; the text intent and the text slot are used to reflect the service required by the text data.

[0128] In this application, the defined slot value can refer to the legal value configured for the defined slot (the legal value can specifically refer to a string). For example, for the defined slot <most suitable population>, the strings "couple dating", "friend outing", and "family parent-child" can be configured as its defined slot values; for the defined slot <scenic spot type>, the defined slot values "5A-level type", "winter outing type", and "natural scenery type" can be configured. For the defined slot values of each defined slot, they can be configured based on specific business requirements, and no further examples will be given here.

[0129] For each defined slot, one or more defined slot values can correspond to it. These defined slot values can form a defined slot value list, and all the defined slot value lists can form a defined set of slot value lists. In short, a defined slot value list in the defined set of slot value lists can correspond to a defined slot, and it will contain one or more defined slot values.

[0130] In the specific implementation of this application, after identifying the text intention of the text data, the text slots of the text data under this text intention can be identified. When identifying the text slots of the text data, not only the defined slot list corresponding to the text intention will be used, but also the set of defined slot value lists will be combined to jointly identify the text slots, so as to accurately identify the slots included in the text data. The specific process of combining the set of defined slot value lists to identify text slots can at least include two parts: 1) First, perform slot matching and identification on the text data through the defined slot list corresponding to the text intention. Thus, in the defined slot list corresponding to the text intention, the defined slots that the text data can match (that is, the possible slots that can be matched, which can be called predicted slots) can be filtered out. Among them, this step can be implemented through the above-mentioned language understanding model (such as the LLM large language model). The model can, based on its own natural language understanding ability, combine the defined slot list to perform semantic understanding and analysis on the text data, so as to match the slots included in the text data (that is, the predicted slots) from the defined slot list; 2) Obtain the defined slot value list corresponding to this predicted slot from the set of defined slot value lists, and then detect whether the predicted slot filtered out in the previous step meets the requirements (that is, whether the text data can match this slot) through this defined slot value list. In other words, the defined slot value list can be used to perform a matching detection on the predicted slot filtered out in the previous step to filter out the predicted slots that do not match (that is, do not meet the requirements). The remaining predicted slots after the final filtering are the final text slots of the text data. That is to say, the specific implementation process of this application embodiment for identifying the text slots of text data through the set of defined slot value lists and the defined slot list corresponding to the text intention includes but is not limited to: First, call a model with natural language understanding ability to perform semantic understanding and analysis on the text data, and initially filter out the slots that the text data matches from the defined slot list. The slots output by the model can be used as predicted slots; then, the defined slot value list configured for the predicted slots can be used to perform a fine detection on each predicted slot to filter out the predicted slots that are not accurately identified by the model.

[0131] It should be noted that there may be more than one text intention of the text data identified in this application. For multiple text intentions, the corresponding predicted slots can be identified one by one according to the above method, and then the defined slot value list of the corresponding predicted slots can be used to perform a matching detection on the corresponding predicted slots to filter out the predicted slots that do not meet the requirements.

[0132] Please refer to Figure 4 , Figure 4 which is a schematic diagram of a brief scenario for text slot identification provided by an embodiment of this application. As Figure 4As shown, assume that the text data to be recognized is "A friend came to City A to find me, and I'm going to take her to stroll around Ancient Town B". Based on the above Figure 3 corresponding embodiment, the text intention of the recognized text data is [Tourist Attraction]. On this basis, first, a model with natural language understanding ability (such as the LLM large language model) needs to obtain the definition slot list corresponding to this text intention [Tourist Attraction] (here it is assumed to include <Region>, <Attraction Type>, <Most Suitable Population>, <Consumption>, <Whether Directly Accessible by Subway>, <Ticket Price>, <Rating>, <Opening Hours>). Through this definition slot list, slot matching recognition can be performed on the text data to screen out the slots that the text data can match from this definition slot list (i.e., predicted slots). As shown in Figure 4 , the possible slots (i.e., predicted slots) obtained by matching prediction include <Region>, <Attraction Type>, <Most Suitable Population>. Among them, the prediction reason for <Region> is that "the user mentioned that a friend came to find him in City A, and City A may meet the region"; the prediction reason for <Attraction Type> is that "the user clearly stated that he would take his friend to stroll around Ancient Town B, and Ancient Town B may meet the attraction type"; the prediction reason for <Most Suitable Population> is that "the user mentioned taking his friend to travel, and the friend may meet the most suitable population".

[0133] Furthermore, for the three predicted slots screened out above, matching detection can be performed through the set of definition slot value lists. Specifically, the definition slot value list corresponding to <Region> can be obtained from the set of definition slot value lists, and then the matching detection of the predicted slot <Region> can be performed through the definition slot value list; the definition slot value list corresponding to <Attraction Type> can also be obtained from the set of definition slot value lists, and then the matching detection of the predicted slot <Attraction Type> can be performed through the definition slot value list; the definition slot value list corresponding to <Most Suitable Population> can also be obtained from the set of definition slot value lists, and then the matching detection of the predicted slot <Most Suitable Population> can be performed through the definition slot value list. It should be noted that only the predicted slots that pass the matching detection (i.e., the matching detection result is a pass result) can be used as a text slot under the corresponding text intention; while the predicted slots that do not pass the matching detection (i.e., the matching detection result is a non-pass result) will be filtered and deleted. Identifying predicted slots through the definition slot list is equivalent to a preliminary slot screening process, and performing matching detection on the predicted slots through the definition slot value list corresponding to the predicted slots is equivalent to a fine filtering process for the predicted slots screened out above. Through preliminary screening and fine filtering, the matching recognition accuracy of the text slots of the text data can be improved.

[0134] It should be noted that after determining the final text intention and text slots of the text data, the true intention indicated by the text data can be determined through the text intention and text slots, and this true intention can reflect the service required by the text data (i.e., the true demand service of the user).

[0135] In a specific implementation, after identifying the text intention and text slots of the text data, the slot value corresponding to the text slot can be obtained from the text data, and then the true intention indicated by the text data can be determined based on the slot value and text intention analysis. For example, assume that the text data to be identified is "Query the weather in City A today", its text intention is [Weather], and the text slots are <Time> and <Location>. In this text data, the slot value corresponding to the text slot <Time> is "today", and the slot value corresponding to the text slot <Location> is "City A". Then, combining the text intention [Weather], the slot value "today", and the slot value "City A", the true intention indicated by the text data can be analyzed and determined as "Obtain the weather conditions in City A today", that is, the service required by the text data (the true demand of the user) is "Obtain the weather in City A today". It should be noted that the text intention of the text data in this application can refer to a broad field or direction, while the true intention of the text data refers to the true demand in that field or direction.

[0136] It should be understood that after identifying the true intention expressed by the text data, corresponding services can be provided to the object of the text (i.e., the object of the text data, such as the user who inputs the text data) according to the true intention of the text (i.e., providing the services required by the user). For ease of understanding, the following will take the text data to be recognized as "How should I move from point A to point B" as an example to illustrate a scenario of providing services to the user based on the true intention. By using the above method for intention recognition, the true intention of the recognized text data can be a navigation intention (i.e., the intention to obtain the navigation path from point A to point B. Here, point A can be used as the starting position point, and point B can be used as the destination position point. Then, this navigation intention is the intention to obtain the navigation path from the starting position point to the destination position point). In this case, for the specific implementation of providing services to the object of the text according to the true intention indicated by the text data, it includes but is not limited to: First, based on the starting position point and the destination position point, different navigation paths from the starting position point to the destination position point can be generated, and these navigation paths can form a navigation path set; for each navigation path in the navigation path set, a travel time analysis can be performed (that is, analyzing the time required to move from the starting position point to the destination position point. For example, different travel times can be obtained based on different movement methods. For example, for the same navigation path, the travel time by car is much less than that by walking), to obtain the path travel time required for each navigation path; then, service indication information (equivalent to navigation information) including the navigation path set and the path travel time required for each navigation path can be generated, and the computer device can send this service indication information to the terminal, and the terminal can push or display this service indication information to the object of the text. Then, the service indication information viewed by the object of the text can specifically include each navigation path, and each navigation path will correspond to a path travel time. The object of the text can move to the destination position point according to the navigation path with the minimum path travel time.

[0137] In summary, in the embodiment of the present application, defined slots are introduced in the text intention recognition of text data. By defining the defined slots corresponding to the intention, the text intention recognition can be divided into two independent atomic tasks (intention preliminary screening task and intention fine filtering task). Through task decomposition, the computational complexity of data reasoning can be reduced in each subdivided atomic task, thereby improving the recognition efficiency. At the same time, by introducing the information of the defined slots dimension to perform fine filtering detection on the intention, the intentions that do not meet the requirements screened out initially can be filtered, and the text intention of the text data can be obtained more accurately. Similarly, in the text slot recognition of text data, defined slot values are introduced. By defining the defined slot values corresponding to the slots, the text slot recognition can be divided into two independent atomic tasks (slot preliminary screening task and slot fine filtering task). Through task decomposition, the computational complexity of data reasoning can be reduced in each subdivided atomic task, thereby improving the recognition efficiency. At the same time, by introducing the information of the defined slot values dimension to perform fine filtering detection on the slots, the slots that do not meet the requirements screened out initially can be filtered, and the text slots of the text data can be obtained more accurately. When both the text intention and text slots of the text data have high accuracy, the true intention determined according to the text intention and text slots can also be more accurate. In addition, since the present application can directly use the defined slots to detect and filter the text intention, and use the defined slot values to detect and filter the text slots, even if the intention recognition scheme provided by the present application is deployed in the model, there is no need to configure dialogue data to train the model. Only by configuring the intention, slot, and slot value can the model be applied to the intention recognition scenario. Through the natural language understanding ability of the model, intention prediction and slot prediction can be performed, and then detection and filtering can be performed through the defined slots and slot values, which can well reduce the data configuration threshold and improve the intention recognition efficiency and real-time performance.

[0138] Based on the above Figure 2 corresponding embodiment, a simple introduction to the text intention recognition of text data is given. Next, the detailed process of obtaining the text intention of the text data by recognizing the intention of the text data will be introduced. Please refer to Figure 5 , Figure 5 is a schematic flowchart of a text intention recognition provided by an embodiment of the present application. This process can correspond to the above Figure 2 corresponding embodiment, for the specific process of performing intention recognition on text data through the defined intention list and the defined slot list set to obtain the text intention of the text data.

[0139] As Figure 5 shown, this process can at least include the following steps S501 - step S504:

[0140] Step S501, perform intent semantic recognition on the text data through a defined intent list to obtain the predicted intent of the text data; the defined intent list includes the predicted intent.

[0141] Specifically, in step S501, after obtaining the text data to be recognized, a language understanding model can be called. The language understanding model can perform semantic analysis on the text data to analyze one or more most likely predicted intents of the text data from the given defined intent list. In a specific implementation, historical text data having a contextually coherent association relationship with the text data can be obtained. Here, the contextually coherent association relationship can refer to a logical association relationship, a dialogue connection association relationship, etc. For example, in a round of conversation, the historical text data sent by the user before inputting the text data has a dialogue connection association relationship with the text data, then it can be considered that there is a contextually coherent association relationship between the historical text data and the text data; then, an intent recognition prompt message can be jointly generated according to the defined intent list, the historical text data and the text data, and the intent recognition prompt message is input into the language understanding model; in the language understanding model, the language understanding model can perform intent semantic recognition on the text data based on its own natural language understanding ability and based on the intent recognition prompt message (combining the defined intent list and the historical text data to perform semantic understanding and analysis on the text data) to identify the intent reflected by the text data (i.e., the predicted intent) from the defined intent list. It should be understood that through intent semantic recognition, the language understanding model can output an intent recognition result, and the intent recognition result will include the predicted intent recognized from the defined intent list and the model's reasoning explanation for the predicted intent (the reasoning explanation is the reason why the model reasons and recognizes the predicted intent). After the model outputs the intent recognition result, the predicted intent included in the intent recognition result can be used as the predicted intent of the text data. For example, in the Figure 3 corresponding exemplary scenario, for the text data "A friend comes to City A to find me and I'm going to show her around Ancient Town B", the model can output an intent recognition result that includes the predicted intents [Tourist Attraction] and [Restaurant], and the reasoning explanation (i.e., the prediction reason) corresponding to each predicted intent. For example, the prediction reason for [Tourist Attraction] is "The user mentioned taking a friend to visit Ancient Town B, and the user's intent is to find tourist attractions".

[0142] Exemplarily, the language understanding ability of the language understanding model itself includes: the model can first perform semantic analysis on the text data to extract the text semantic representation corresponding to the text (the text semantic representation is used to represent the comprehensive semantic information of the text data); then, the text semantic representation can be used to calculate the attribution probability of the text data belonging to each defined intention in the defined intention list, and the value of the attribution probability corresponding to any defined intention is used to represent the degree to which the intention expressed by the text data is the same as the defined intention, that is, the attribution probability of the text data belonging to each defined intention is used to represent the degree to which the indicated expression intention of the text data is the same as each defined intention; based on this, the defined intention in the defined intention list with an attribution probability greater than the probability threshold (the probability threshold can be set manually, for example, 0.8, 0.85, etc.) can be selected and determined as the predicted intention of the text data. Since the predicted intention is a certain defined intention in the defined intention list, the defined intention list will contain the predicted intention. It should be noted that there may not be only one attribution probability greater than the probability threshold here, so there may not be only one predicted intention here. Of course, there may also be 0 predicted intentions here. When there is no attribution probability greater than the probability threshold, the defined intention with the largest attribution probability value can be used as the predicted intention.

[0143] It should be noted that in the intelligent dialogue scenario, the multiple text data used for the front and back inputs may be coherent. Then, when predicting and identifying the intention of the text data here, the historical text data of the text data in a round of dialogue (the historical text data has a context coherence relationship with the text data) can also be obtained. Then, the text semantic representation of the historical text data is obtained, and by combining the text semantic representation of the historical text data with the text semantic representation of the current text data, the predicted intention of the current text data can be screened out from the defined intention list. Since there is a context coherence relationship between the historical text data and the text data, by combining the text semantic representation of the historical text data, the predicted intention of the current text data can be analyzed more accurately.

[0144] It should also be noted that the intention recognition prompt information generated by this application only contains a defined intention list in addition to the historical text data and the text data expressed by the current user. That is to say, this application only inputs this one reference information, the defined intention list, into the language understanding model as the reasoning reference / basis for the language understanding model. Compared with the traditional method of inputting a large amount of more abundant information (such as defined intentions, intention description texts of defined intentions, etc.) into the model, this application can input as many defined intentions as possible into the language understanding model. For example, when there is a limit on the input length of the model, since the traditional model input content also includes intention description texts of defined intentions, and these intention description texts are usually long, then the number of defined intentions that can be input into the model will be less; while in this application, a large number of defined intentions can be put into the language understanding model, and there is no restriction on the input quantity of defined intentions.

[0145] Step S502, obtain the defined slot list corresponding to the predicted intention from the set of defined slot lists.

[0146] Specifically, for the above-identified predicted intentions, it can be understood that they are intentions initially screened from the defined intention list based on the basic natural language understanding ability. For these predicted intentions, this application uses the defined slot list under the predicted intention to perform an adaptability detection on them to test their adaptability to the text data. Through the detection, the predicted intentions that do not meet the requirements (that is, are not adaptable and are not accurately screened out) can be filtered to obtain sufficiently adaptable and accurate predicted intentions, and use them as the final text intention of the text data. Then, after the above-identified predicted intentions, for each predicted intention, the corresponding defined slot list can be obtained from the set of defined slot lists.

[0147] Step S503, perform an intention adaptability detection on the predicted intention through the defined slot list corresponding to the predicted intention, and obtain the adaptability detection result of the predicted intention.

[0148] In a specific implementation, for the convenience of distinction, the list of defined slot positions corresponding to the predicted intent can be first referred to as the first target slot position list; then, the intent description information configured for the predicted intent and the above historical text data can be obtained; it should be understood that after each defined intent is configured in this application, corresponding description information can be configured for each defined intent as the intent description information of the defined intent, and the intent description information of a defined intent can be used to accurately describe what the defined intent needs to do (i.e., the functions possessed by the defined intent). For example, for the defined intent of [weather], which is mainly used for weather query, the intent description information of [weather] can be "weather query"; another example is that for the defined intent of [airplane], its functions include "flight ticket reservation", "flight query", etc., then for [airplane], its intent description information can include "flight ticket reservation", "flight query", etc. Based on this, after determining the predicted intent of the text data in this application, the intent description information of the predicted intent can be obtained. After that, intent detection prompt information can be generated according to the predicted intent, the intent description information of the predicted intent, the first target slot position list, the historical text data and the text data, and the intent detection prompt information can be input into the language understanding model. In the language understanding model, the adaptability between the text data and the predicted intent can be analyzed based on the intent detection prompt information to obtain the adaptability detection result of the predicted intent. That is to say, the specific implementation process of this application embodiment for identifying the text intent of text data through the defined intent list and the defined slot position list set includes but is not limited to: First, by calling a model with natural language understanding ability, semantic understanding and analysis of the text data is performed, and the intent reflected by the text data is preliminarily screened out from the defined intent list, and the intent output by the model can be used as the predicted intent; then, through the defined slot position list configured under the predicted intent, each predicted intent can be finely detected to filter out the predicted intents with inaccurate model predictions.

[0149] For example, taking the above Figure 3Taking the corresponding embodiment as an example, after obtaining the current user's expression "A friend is coming to City A to visit me, and I'm going to show her around Ancient Town B", it can be used as the text data to be recognized. For this text data, which is the first sentence input by the user in the current dialogue scenario, it does not contain historical text data. Then, a defined intent list can be obtained as: [Tourist Attractions], [Restaurants], [Hotels], [Computers], [Trains], [Weather], [Airplanes]. Then, based on this defined intent list, historical text data, and the current text data to be recognized, the embodiment of the present application can generate an intent recognition prompt message. For example, this intent recognition prompt message is: "You are a dialogue analyzer. Here is a defined intent list for you: [Tourist Attractions], [Restaurants], [Hotels], [Computers], [Trains], [Weather], [Airplanes]. The current user's expression is: A friend is coming to City A to visit me, and I'm going to show her around Ancient Town B; Please give the two most likely predicted intents (without repetition) from the above-defined intent list in a specific format (such as JSON format)". Through this intent recognition prompt message, it can be used to instruct the dialogue analyzer to select the two most likely predicted intents for the text data to be recognized (the current user's expression) from the given defined intent list. Then, this intent recognition prompt message can be input into a language understanding model with natural language understanding capabilities (such as an LLM model). This language understanding model is the dialogue analyzer. Through the language understanding model, it can combine historical text data to understand the text data to be recognized, so as to analyze the predicted intent of the current text data from the defined intent list. As Figure 3 shown, assume that the possible intents (i.e., predicted intents) recognized and predicted by the model include [Tourist Attractions] and [Restaurants]. Among them, the reason for the model's prediction of [Tourist Attractions] is that "the user mentioned taking a friend to visit Ancient Town B, and the user's intention is to find tourist attractions"; the reason for the prediction of [Hotels] is that "the user mentioned that a friend is coming to visit him in City A, and accommodation may be needed, so the intention is to find a hotel".

[0150] Furthermore, for the two predicted intents screened out above, adaptability detection can be performed by defining a set of slot lists. Taking the adaptability detection of the predicted intent [Tourist Attraction] through defining a slot list as an example, the defined slot list corresponding to the predicted intent [Tourist Attraction] includes: <Region>, <Attraction Type>, <Most Suitable Population>, <Consumption>, <Whether Directly Accessible by Subway>, <Ticket Price>, <Rating>, <Opening Hours>. This defined slot list can be called the first target slot list. Then, based on this predicted intent (i.e., the intent name [Tourist Attraction]), the intent description information of the predicted intent (Tourist attraction query), this first target slot list, this historical text data, and this text data, an intent detection prompt message can be generated. This intent detection prompt message is, for example: "You are a dialogue analyzer, and there is an intent as follows {Intent name: Tourist Attraction; Intent description information: Tourist attraction query; Slots under the intent include: <Region>, <Attraction Type>, <Most Suitable Population>, <Consumption>, <Whether Directly Accessible by Subway>, <Ticket Price>, <Rating>, <Opening Hours>}; The current user's expression is: A friend came to City A to find me, and I'm going to take her to visit Ancient Town B; Please indicate in a specific format (such as JSON format) whether the current user's expression belongs to this intent". Through this intent detection prompt message, it can be used to instruct the dialogue analyzer to detect whether the current user's expression belongs to the predicted intent [Tourist Attraction] based on the given first target slot list (i.e., to detect whether the predicted intent [Tourist Attraction] is adaptable to this text data). Then, this intent detection prompt message can be input into the language understanding model. Through the language understanding model, it can combine the historical text data and the first target slot list to understand the text data in order to analyze whether the current text data belongs to this predicted intent. For example, assume the model analyzes that this text data belongs to this predicted intent because: In the current user's expression, it is clearly mentioned that they are going to take their friend to a water town ancient town, and the water town ancient town is associated with the slot <Attraction Type> in tourist attractions.

[0151] It can be understood that when a language understanding model performs an adaptability detection on a certain predicted intention, it can first combine historical text data to analyze whether the text data contains text information associated with a certain defined slot in the first target slot list, that is, to determine whether the text data contains a string that matches a certain defined slot through the context of the text data; the string that matches a certain defined slot can be understood as the slot value of the defined slot matched in the text data. Based on this, in the case where there is no slot value that matches any defined slot in the text data, it can be considered that the text data does not involve any content related to a certain defined slot under any associated predicted intention, and the text data does not care about any content (i.e., slots) covered by the predicted intention. And in the case where the text data does not involve any content related to the predicted intention, it can be considered that the adaptability detection result of the identified predicted intention is a non-passing result of the adaptability detection (this non-passing result of the adaptability detection can indicate that the predicted intention is not adapted to the text data); if it is determined that there is a slot value that matches a certain defined slot in the text data, in the case where there is a slot value of a certain defined slot in the text data, it can be considered that there is a certain content associated with the predicted intention in the text data. In this way, it is determined that the adaptability detection result of the predicted intention is a passing result of the adaptability detection (this passing result of the adaptability detection can indicate that the predicted intention is adapted to the text data).

[0152] In summary, when a language understanding model performs an adaptability detection on a certain predicted intention, it can first perform a semantic analysis on the current text data based on its own language understanding ability and in combination with historical text data to analyze whether the text data contains text information associated with a certain defined slot in the first target slot list (the defined slot list corresponding to the predicted intention). If it contains text information associated with a certain defined slot, then it can be considered that the predicted intention is adapted to the text data; otherwise, the predicted intention is not adapted and needs to be filtered and deleted.

[0153] Step S504, if it is determined that the adaptability detection result indicates that the predicted intention is adapted to the text data, then the predicted intention is determined as the text intention of the text data.

[0154] Specifically, as can be seen from the above, for a certain predicted intention, the corresponding adaptability detection result can include a passing result of the adaptability detection and a failing result of the adaptability detection. The passing result of the adaptability detection can indicate that the predicted intention is adaptable to the text data, while the different result of the adaptability detection can indicate that the predicted intention is not adaptable to the text data. Then, based on this, for a certain predicted intention, if it is determined that its adaptability detection result indicates that the predicted intention is not adaptable to the text data, it can be filtered and deleted. Only the remaining predicted intentions after filtering can be used as a text intention of the text data. In other words, only the predicted intention whose adaptability detection result indicates that the predicted intention is adaptable to the text data can be determined as a text intention of the text data.

[0155] In summary, in the embodiment of the present application, a defined slot is introduced in the text intention recognition of text data. By defining the defined slot corresponding to the intention, the text intention recognition can be divided into two independent atomic tasks (intention preliminary screening task and intention fine filtering task). Through task decomposition, the computational complexity of data inference can be reduced in each subdivided atomic task, thereby improving the recognition efficiency. At the same time, by introducing the information of this dimension of the defined slot to perform fine filtering detection on the intention, the intentions that do not meet the requirements screened out initially can be filtered, and the text intention of the text data can be obtained more accurately. The true intention of the text data obtained based on the text intention with high accuracy can also have higher precision. It should be understood that the present application can use the intention output by the model as a preliminary predicted intention. Subsequently, the present application can adopt the method of introducing a defined slot list to perform fine detection and filtering on the predicted intention to detect and filter the output result of the model, and relatively accurate recognition results can be obtained without training the model, which can greatly reduce the cost of configuring the model training sample data.

[0156] Based on the above Figure 2 In the corresponding embodiment, a simple introduction to the text slot recognition of text data is given. Next, the detailed process of obtaining the text slot of the text data by recognizing the slot of the text data will be introduced. Please refer to Figure 6 , Figure 6 is a schematic flowchart of a process for performing text slot recognition provided by an embodiment of the present application. This process can correspond to the above Figure 2 In the corresponding embodiment, for the specific process of performing slot recognition on text data through a defined slot value list set and a defined slot list corresponding to the text intention to obtain the text slot of the text data. As Figure 6 shown, this process can at least include the following steps S601 - step S605:

[0157] Step S601, determine the defined slot list corresponding to the text intention as the second target slot list.

[0158] Specifically, from the above description, it can be seen that the number of predicted intents recognized for the text data may be more than one (including one or more). After performing an adaptability detection on the predicted intents, the number of remaining predicted intents after detection and filtering may also be one or more. Then, the text intent of the text data may also be one or more. In an actual scenario, if multiple predicted intents all pass the adaptability detection, a certain predicted intent can be selected from the multiple predicted intents that have passed the adaptability detection as the final text intent of the text data. For example, the most suitable predicted intent can be selected from the predicted intents as the final text intent of the text data. For example, if the text data has the most associated information with a certain predicted intent, then the predicted intent with the most associated information can be determined to be the most suitable for the text data, and it can be used as the final text intent.

[0159] In a specific implementation, the corresponding definition slot list for this text intent can be obtained from the defined slot list set, and for the sake of distinction, it can be called the second target slot list.

[0160] Step S602: Perform slot matching recognition on the text data through the second target slot list to obtain the predicted slots of the text data; the second target slot list contains the predicted slots.

[0161] In a specific implementation, the above-mentioned language understanding model can be called to perform slot matching recognition on the text data to obtain the predicted slots of the text data. The specific implementation process can include but is not limited to: First, the intent description information configured for this text intent and the historical text data that has a context coherent association relationship with this text data can be obtained; then, according to the text intent, the intent description information of the text intent, the second target slot list, the historical text data, and the text data, a slot matching prompt information can be generated. The slot matching prompt information can be input into the language understanding model. In this language understanding model, based on its own natural language understanding ability, slot matching recognition can be performed on the text data based on the slot matching prompt information (combining the second target slot list and the historical text data to perform semantic understanding and analysis on the text data) to match and recognize the predicted slots of the text data from the second target slot list. It should be understood that in the language understanding model, when the model performs slot matching recognition on the text data based on the slot matching prompt information, a slot recognition result can be output. This slot recognition result includes the predicted slots matched and recognized by the language understanding model from the second slot list, the text slot values in the text data that match the predicted slots, and the reasoning description corresponding to the predicted slots (that is, the reason why the model recognizes this predicted slot). After obtaining the slot recognition result output by the model, the predicted slots included in this slot recognition result can be used as the predicted slots of the text data. For example, in the above Figure 4In the corresponding exemplary scenario, for the text data "A friend came to City A to visit me, and I plan to show her around Ancient Town B", the model can output a slot recognition result. This slot recognition result contains the predicted intents <Region>, <Scenic Spot Type>, and <Most Suitable Population>, the text slot values in the text data that match the predicted intents (i.e., the strings in the text data. For example, in the text data, the text slot value that matches the predicted intent <Region> is 'City A', the text slot value that matches the predicted intent <Scenic Spot Type> is 'Ancient Town B', and the text slot value that matches the predicted intent <Most Suitable Population> is 'friend'), and also contains the reasoning explanation for each predicted intent (i.e., the reason for prediction). For example, the reason for predicting <Region> is "The user mentioned that a friend came to City A to visit him, so City A may match the region". It can be seen that the reasoning explanation (reason for prediction) for the predicted slot is determined based on the corresponding text slot value in the text data.

[0162] Exemplarily, the text data can be understood as a character sequence composed of one or more characters. Then, the model can analyze whether the character sequence contains a string that matches a certain defined slot in the second target slot list, and there is a similarity between the character semantics of this string and the slot semantics of this defined slot; if there is a string in the text data that matches a certain defined slot, then this defined slot can be used as the predicted slot matched by the model. It should be noted that for a certain defined slot, the string that matches it in one or more characters can be understood as the text slot value in the text data that matches this defined slot.

[0163] It should be understood that since the predicted slot is determined from the second target slot list, the second target slot list contains this predicted slot.

[0164] Step S603, obtain the list of defined slot values corresponding to the predicted slot from the set of defined slot value lists.

[0165] In specific implementation, there may be more than one predicted slot determined above, and the number may be one or more. For any predicted slot, its matching detection can be performed through the corresponding list of defined slot values. Then, for a certain predicted slot, it is necessary to first obtain the list of defined slot values corresponding to this predicted slot from the set of defined slot value lists.

[0166] Step S604, perform a matching detection on the predicted slot through the list of defined slot values corresponding to the predicted slot, and obtain the matching detection result of the predicted slot.

[0167] In specific implementation, a language understanding model can be used to perform a matching detection on a predicted slot by predicting a list of defined slot values corresponding to the slot, and the specific implementation process includes but is not limited to: First, historical text data that has a contextually coherent association with the text data can be obtained; based on the text intention, the predicted slot under the text intention, the list of defined slot values corresponding to the predicted slot, the text slot value in the text data that matches the predicted slot, the reasoning description corresponding to the predicted slot, the historical text data, and the text data, a slot detection prompt message can be generated, and the slot detection prompt message can be input into the language understanding model. In the language understanding model, the matching between the text data and the predicted slot can be detected based on the slot detection prompt message to obtain the matching detection result of the predicted slot.

[0168] In a language understanding model, a defined slot value list can be combined to determine whether the text slot value in the text data actually belongs to the predicted slot (that is, whether it matches the predicted slot. If it is verified that the text slot value belongs to the predicted slot, it can be considered that the predicted slot matches the text data). Specifically, for the convenience of distinction, the text slot value in the above-mentioned text data output by the model that matches the predicted slot can be first determined as the verification slot value of the predicted slot included in the text data (that is, the text slot value of the predicted slot included in the text data is called the verification slot value); further, for the convenience of distinction, the defined slot value list corresponding to the predicted slot can be determined as the target slot value list (that is, it is called the target slot value list), and a similarity analysis is performed on the verification slot value and the target slot value list to obtain the similarity analysis result between the verification slot value and the target slot value list; among them, the similarity analysis result here can include the similarity existence result and the similarity non-existence result. The similarity existence result can be used to indicate that there is a similarity between the verification slot value and the target slot value list, and the similarity non-existence result can be used to indicate that there is no similarity between the verification slot value and the target slot value list. If the similarity analysis result indicates that there is a similarity between the verification slot value and the target slot value list, it can be determined that there is a very high similarity or correlation between the verification slot value and a certain defined slot value or some defined slot values in the target slot value list. In this case, it can be considered that some defined slot values under the predicted slot are hit in the text data, and the verification slot value in the text data is strongly related to the predicted slot, and the matching detection result of the predicted slot can be determined as a matching detection pass result; if the similarity analysis result indicates that there is no similarity between the verification slot value and the target slot value list, it can be determined that there is no similarity or correlation between the verification slot value and any defined slot value in the target slot value list, and the verification slot value does not hit any defined slot value under the predicted slot. In this case, it can be considered that the verification slot value in the text data is not related to the predicted slot, and the matching detection result of the predicted slot can be determined as a matching detection failure result.

[0169] As can be seen from the above, to perform a similarity analysis on the verification slot value and the target slot value list to obtain the similarity analysis result between the verification slot value and the target slot value list, it can be determined by calculating the similarity or relevance between the verification slot value and each defined slot value in the target slot value list. If it is determined that there is a certain defined slot value in the target slot value list that has similarity with the verification slot value, then it can be considered that the similarity analysis result between the verification slot value and the target slot value list is a similarity existence result; and if it is determined that there is no defined slot value in the target slot value list that has similarity with the verification slot value, then it can be determined that the similarity analysis result between the verification slot value and the target slot value list is a similarity non-existence result.

[0170] In a specific implementation, the specific implementation process of performing a similarity analysis on the verification slot value and the target slot value list to obtain the similarity analysis result between the verification slot value and the target slot value list may include, but is not limited to: First, the word similarity between the verification slot value and each defined slot value in the target slot value list can be calculated, and the word similarities corresponding to the respective defined slot values can form a word similarity set; further, the word similarity set can be traversed to query whether there is a word similarity greater than the similarity threshold (which can be set manually, for example, the threshold can be 0.95, 0.87, 0.88, etc.) in the word similarity set; if it is traversed and determined that there is a word similarity greater than the similarity threshold in the word similarity set, then it can be determined that there is a defined slot value in the target slot value list that is similar to the verification slot value. In this case, the similarity analysis result between the verification slot value and the target slot value list can be determined as a similarity existence result; and if there is no word similarity greater than the similarity threshold in the word similarity set, then it can be determined that there is no defined slot value in the target slot value list that is similar to the verification slot value. In this case, the similarity analysis result between the verification slot value and the target slot value list can be determined as a similarity non-existence result.

[0171] It should be noted that both the verification slot value and a certain defined slot value can be understood as a text word. Then, calculating the similarity between the verification slot value and a certain defined slot value can be understood as calculating the word similarity between two text words. For the specific implementation method of calculating the word similarity between two text words, any method that can calculate the similarity between two text words can be adopted in this application. For example, methods such as calculating vector distance, calculating cosine similarity, and analyzing similarity based on a statistical model can be used. For the sake of understanding, the following will illustrate the method of determining the similarity between the verification slot value and a certain defined slot value by calculating the vector distance to calculate the similarity between two text words.

[0172] Taking the case where the target slot value list contains the defined slot value S i (where i is a positive integer) as an example, the word similarity set will contain the word similarity between the verification slot value and the defined slot value S i For calculating the word similarity between the verification slot value and each defined slot value in the target slot value list to obtain the word similarity set, it is equivalent to using the method of calculating the word similarity between the verification slot value and the defined slot value S i to calculate the word similarity between the verification slot value and each defined slot value, and thus the word similarity set can be obtained. In a specific implementation, for calculating the word similarity between the verification slot value and the defined slot value S i The specific implementation process includes but is not limited to: First, it is necessary to obtain the vector feature for characterizing the verification slot value (which can be called the first word vector), and the vector feature for characterizing the defined slot value S i (which can be called the second word vector); further, the vector distance between the first word vector and the second word vector can be calculated; according to the vector distance, the word similarity between the verification slot value and the defined slot value S i can be determined. For example, the present application can pre-configure different distance intervals, and thus a configuration distance interval set containing different configured distance intervals can be obtained. For each configured distance interval, a similarity can be configured as the similarity having a mapping relationship with it. In this way, each configured distance interval can have a mapping relationship with a similarity. The present application can construct a distance mapping table based on the mapping relationship between each configured distance interval in the configuration distance interval set and a similarity in the similarity set, and store the distance mapping table. Then, after obtaining the vector distance between the first word vector and the second word vector as described above, the distance mapping table can be obtained; further, the configured distance interval where the vector distance is located can be obtained from the configuration distance interval set, and the configured distance interval where the vector distance is located can be determined as the target configured distance interval (i.e., called the target configured distance interval); then, the similarity in the similarity set that has a mapping relationship with the target configured distance interval can be determined as the word similarity between the verification slot value and the defined slot value S i

[0173] Optionally, it can be understood that since the language understanding model (LLM model) in this application has emergent capabilities, adding the reasoning explanation of the predicted slot (that is, the thinking process of the model for the predicted slot) to the slot detection prompt information and inputting it into the language understanding model can serve as a reference for the language understanding model, thereby improving the final reasoning ability of the language understanding model, and further improving the accuracy of the model output result. However, in actual application scenarios, it can be selected whether to input the reasoning explanation of the predicted slot into the language understanding model according to specific requirements. In other words, when constructing the slot detection prompt information, the reasoning explanation of the predicted slot can be not used as a component of the slot detection prompt information.

[0174] Step S605, if it is determined that the matching detection result indicates that the predicted slot matches the text data, then determine the predicted slot as the text slot of the text data.

[0175] Specifically, through the above steps, it can be known that the matching detection result can include a matching detection pass result and a matching detection fail result. For a certain predicted slot, if its matching detection result is a matching detection pass result, it can be determined that the predicted slot matches the text data, and the predicted slot can be determined as the final text slot of the text data; otherwise, it can be determined that the predicted slot does not match the text data, and the predicted slot needs to be filtered and deleted to filter out the non-matching predicted slots and retain the matching predicted slots.

[0176] For example, taking the above Figure 4 corresponding embodiment as an example, as Figure 4 shown, assume that the text data to be recognized is "A friend came to City A to find me, and I'm going to take her to stroll around Ancient Town B". Based on the above Figure 3According to the corresponding embodiment, the text intention of the recognized text data is [tourist attraction]. On this basis, the intention description information of this text intention (which is "tourist attraction query") and the historical text data of this text data (there is no historical text data, which is a null value) can be obtained first. Then, the defined slot list corresponding to this text intention [tourist attraction] can be obtained (here it is assumed to include <region>, <attraction type>, <most suitable group of people>, <consumption>, <whether directly accessible by subway>, <ticket price>, <rating>, <opening hours>), and this defined slot list can be called the second target slot list. According to this text intention (that is, the name of the text intention [tourist attraction]), the intention description information of the text intention (tourist attraction query), this second target slot list and this text data, a slot matching prompt message can be generated. For example, this slot matching prompt message is: "You are a dialogue analyzer, and there is an intention named "tourist attraction" as follows: {intention name: tourist attraction; intention description information: tourist attraction query; slots under the intention include: <region>, <attraction type>, <most suitable group of people>, <consumption>, <whether directly accessible by subway>, <ticket price>, <rating>, <opening hours>}; the current user's expression is: A friend comes to City A to find me, and I am going to take her to stroll around Ancient Town B; please give the slots that the user's current expression may contain in a specific format (such as json format), and return as many as possible, even if there is only a weak association". Through this slot matching prompt message, it can be used to instruct the dialogue analyzer to select multiple predicted slots included in the text data from the given second target slot list. Then, this slot matching prompt message can be input into the language understanding model, that is, the dialogue analyzer. Through the language understanding model, the text data can be understood to analyze the predicted slots included in the text data from the second target slot list. As Figure 4 shown, assume that the possible slots (that is, the predicted slots) analyzed and predicted by the model include <region>, <attraction type>, <most suitable group of people>. Among them, the prediction reason for <region> is that "the user mentioned that a friend came to find him in City A, and City A may meet the region"; the prediction reason for <attraction type> is that "the user clearly stated that he is going to take a friend to stroll around Ancient Town B, and Ancient Town B may meet the attraction type"; the prediction reason for <most suitable group of people> is that "the user mentioned that he is going to take a friend to travel, and the friend may meet the most suitable group of people".

[0177] Further, for the three predicted slots screened above, matching detection can be performed by defining a set of slot value lists. Taking the matching detection of the predicted slot <Most suitable population> by defining a slot value list as an example, the defined slot value list corresponding to the predicted slot <Most suitable population> includes: "Couple dating", "Friends traveling", "Family parent-child"; according to the text intention [Tourist attraction], the predicted slot <Most suitable population>, the defined slot value list ("Couple dating", "Friends traveling", "Family parent-child"), the verification slot value "Friends" in the text data, the reasoning explanation of this predicted slot (i.e., the prediction reason "The user mentioned taking friends to travel, and friends may fit the most suitable population") and this text data, a slot detection prompt message can be generated. This slot detection prompt message is, for example: "You are a dialogue analyzer with an intention named 'Tourist attraction', and there is a slot named 'Most suitable population' in it. Its legal values are as follows: 'Couple dating', 'Friends traveling', 'Family parent-child'; the current user's expression is: Friends come to City A to find me, and I'm going to take her to visit Ancient Town B; the following is a possible slot value: {Slot: Most suitable population; Slot value: Friends; Reason: The user mentioned taking friends to the water town ancient town}; please judge whether the following slot value belongs to the <Most suitable population> slot. If it does, which value is it in the list? Return it in a specific format (such as json format)". Through this slot detection prompt message, it can be used to instruct the dialogue analyzer to detect whether the current user's expression contains the slot value of a certain defined slot based on the given defined slot value list (including multiple legal values). For example, when the above model performs slot matching, it analyzes that the predicted slots in the text data include <Most suitable population> because the text data involves a string "Friends" associated with it. Then, the string "Friends" in this text data can be determined as the verification slot value. Through analysis, the model analyzes that the verification slot value "Friends" belongs to the <Most suitable population> slot because the verification slot value has strong similarity with the defined slot value "Friends traveling". Based on this, it can be determined that there is a match between the predicted slot <Most suitable population> and this text data. The matching detection result of this predicted slot is a pass result, and this predicted slot can be used as the final text slot.

[0178] In summary, in the text slot recognition of the embodiments of the present application, a defined slot value is introduced. By means of the defined slot value corresponding to the slot, the text slot recognition can be divided into two independent atomic tasks (slot preliminary screening task and slot fine filtering task). Through task decomposition, the computational complexity of data inference can be reduced in each subdivided atomic task, thereby improving the recognition efficiency. At the same time, by introducing the information of the defined slot value dimension to perform fine filtering detection on the slot, the slots that do not meet the requirements screened out initially can be filtered, and the text slots of the text data can be obtained more accurately. The true intention of the text data obtained based on the text slots with high accuracy can also have higher precision. It should be understood that the present application can use the intention output by the model as the preliminary predicted intention. Subsequently, the present application can adopt the method of introducing a defined slot list to perform fine detection and filtering on the predicted intention to detect and filter the output result of the model, and relatively accurate recognition results can be obtained without training the model, which can greatly reduce the cost of configuring the model training sample data.

[0179] Further, for the convenience of understanding the overall logic of the intention recognition method provided by the embodiments of the present application, please also refer to Figure 7 , Figure 7 which is a schematic diagram of a logical architecture provided by the embodiments of the present application. Among them, as Figure 7 shown, the logical architecture may at least include: a domain preliminary screening component, a domain fine detection component, a domain statistics component, a slot preliminary screening component, a slot fine detection component, and a general reply component. The logical functions implemented by each component will be briefly introduced below.

[0180] Domain preliminary screening component: The domain in the present application can actually refer to the text intention of the text data, that is, the direction that the text data wants to express. Then the domain preliminary screening component here can also be understood as the preliminary screening component of the text intention, which can be used to initially screen out the predicted intention (i.e., the predicted domain, that is, the intention, domain, or direction that the text data may want to express) adapted to the text data from the defined intention list (the defined intentions in the defined intention list are the defined domains).

[0181] Domain fine detection component: The domain fine detection component, that is, the intention fine detection component, can be used to perform adaptability detection on the above-mentioned initially screened predicted intention to filter out the inadaptable predicted intention.

[0182] Domain Statistics Component: The domain statistics component can be used to count the number of remaining predicted intents after filtering by the domain fine detection component. If the number of predicted intents is 0, it can be considered that the text data has not expressed a certain intent, and a general response (such as "Sorry, your needs cannot be determined") can be sent to the user of the input text data by calling the general response component. If it is determined that the number of predicted intents is a non-zero value, it can be considered that the text data expresses a certain defined intent, and slot matching recognition can be performed under this defined intent.

[0183] Slot Initial Screening Component: The domain fine detection component can send the remaining predicted intents after filtering to the slot initial screening component. The slot initial screening component can call the language understanding model to identify the defined slots matched by the text data from the defined slot list corresponding to the predicted intent as the predicted slots of the text data.

[0184] Slot Fine Detection Component: The slot fine detection component can be used to perform matching detection on the predicted slots obtained by the above slot initial screening component to filter out the unmatched predicted slots. The finally remaining predicted slots after filtering can be used as the text slots of the text data. Through the text intent and text slots of the text data, the true intent of the text data can be analyzed and determined.

[0185] For the specific implementation methods of each component in the embodiments of the present application, reference can be made to the corresponding descriptions in the previous embodiments, which will not be elaborated here, and the beneficial effects brought by them will not be elaborated too much.

[0186] Further, please refer to Figure 8 , Figure 8 which is a schematic structural diagram of a data processing device provided by an embodiment of the present application. The data processing device can be a computer program (including program code) running in a computer device. For example, the data processing device is an application software. The data processing device can be used to execute Figure 3 the method shown. As Figure 8 shown, the data processing device 1 can include: a data acquisition module 11, an intent recognition module 12, a slot recognition module 13, and an intent determination module 14.

[0187] The data acquisition module 11 is used to acquire the text data to be recognized;

[0188] The intent recognition module 12 is used to perform intent recognition on the text data through the defined intent list and the set of defined slot lists to obtain the text intent of the text data. There is a one-to-one correspondence between the defined slot lists in the set of defined slot lists and the defined intents in the defined intent list;

[0189] The slot recognition module 13 is used to perform slot recognition on the text data by defining a set of slot value lists corresponding to the text intent and a defined slot list, so as to obtain the text slots of the text data; there is a one-to-one correspondence between the defined slot value lists in the set of defined slot value lists and the defined slots in the defined slot list; the text intent and the text slots are used to reflect the services required by the text data.

[0190] Among them, for the specific implementation manners of the data acquisition module 11, the intent recognition module 12, and the slot recognition module 13, reference can be made to the descriptions of steps S101 - S103 in the corresponding embodiments above, which will not be elaborated here. Figure 2 For the specific implementation manners of the data acquisition module 11, the intent recognition module 12, and the slot recognition module 13, reference can be made to the descriptions of steps S101 - S103 in the corresponding embodiments above, which will not be elaborated here.

[0191] In one embodiment, the specific implementation manner of the intent recognition module 12 for performing intent recognition on the text data by defining an intent list and a set of defined slot lists to obtain the text intent of the text data includes:

[0192] Performing intent semantic recognition on the text data through the defined intent list to obtain the predicted intent of the text data; the defined intent list includes the predicted intent;

[0193] Obtaining the defined slot list corresponding to the predicted intent from the set of defined slot lists;

[0194] Performing intent adaptability detection on the predicted intent through the defined slot list corresponding to the predicted intent to obtain the adaptability detection result of the predicted intent;

[0195] If it is determined that the adaptability detection result indicates that the predicted intent is suitable for the text data, then the predicted intent is determined as the text intent of the text data.

[0196] In one embodiment, the specific implementation manner of the intent recognition module 12 for performing intent semantic recognition on the text data through the defined intent list to obtain the predicted intent of the text data includes:

[0197] Obtaining historical text data that has a contextually coherent association relationship with the text data;

[0198] Generating intent recognition prompt information according to the defined intent list, the historical text data, and the text data, and inputting the intent recognition prompt information into the language understanding model;

[0199] In the language understanding model, performing intent semantic recognition on the text data based on the intent recognition prompt information, and identifying the predicted intent of the text data from the defined intent list.

[0200] In one embodiment, the specific implementation manner of the intent recognition module 12 for performing intent adaptability detection on the predicted intent through the defined slot list corresponding to the predicted intent to obtain the adaptability detection result of the predicted intent includes:

[0201] Determine the list of defined slots corresponding to the predicted intent as the first target slot list;

[0202] Obtain the intent description information configured for the predicted intent and the historical text data that has a contextually coherent association with the text data;

[0203] Generate an intent detection prompt message based on the predicted intent, the intent description information of the predicted intent, the first target slot list, the historical text data, and the text data, and input the intent detection prompt message into the language understanding model;

[0204] In the language understanding model, analyze the suitability between the text data and the predicted intent based on the intent detection prompt message to obtain the suitability detection result of the predicted intent.

[0205] In one embodiment, the slot recognition module 13 performs slot recognition on the text data by defining a set of slot value lists and a list of defined slots corresponding to the text intent. The specific implementation method for obtaining the text slots of the text data includes:

[0206] Determine the list of defined slots corresponding to the text intent as the second target slot list;

[0207] Perform slot matching recognition on the text data through the second target slot list to obtain the predicted slots of the text data; the second target slot list includes the predicted slots;

[0208] Obtain the list of defined slot values corresponding to the predicted slots from the set of defined slot value lists;

[0209] Perform a matching detection on the predicted slots through the list of defined slot values corresponding to the predicted slots to obtain the matching detection result of the predicted slots;

[0210] If it is determined that the matching detection result indicates that the predicted slots match the text data, then determine the predicted slots as the text slots of the text data.

[0211] In one embodiment, the specific implementation method for the slot recognition module 13 to perform slot matching recognition on the text data through the second target slot list to obtain the predicted slots of the text data includes:

[0212] Obtain the intent description information configured for the text intent and the historical text data that has a contextually coherent association with the text data;

[0213] Generate a slot matching prompt message based on the text intent, the intent description information of the text intent, the second target slot list, the historical text data, and the text data, and input the slot matching prompt message into the language understanding model;

[0214] In a language understanding model, slot matching recognition is performed on text data based on slot matching hint information to obtain a slot recognition result; the slot recognition result includes a predicted slot recognized and matched by the language understanding model from a second slot list, a text slot value in the text data that matches the predicted slot, and an inference description corresponding to the predicted slot;

[0215] Determine the predicted slot in the text data as the predicted slot included in the slot recognition result.

[0216] In one embodiment, the specific implementation manner for the slot recognition module 13 to perform a matching detection on the predicted slot through a defined slot value list corresponding to the predicted slot to obtain a matching detection result of the predicted slot includes:

[0217] Obtain historical text data that has a context coherent association relationship with the text data;

[0218] Generate slot detection hint information based on the text intention, the predicted slot under the text intention, the defined slot value list corresponding to the predicted slot, the text slot value in the text data that matches the predicted slot, the inference description corresponding to the predicted slot, the historical text data, and the text data, and input the slot detection hint information into the language understanding model;

[0219] In the language understanding model, detect the matching between the text data and the predicted slot based on the slot detection hint information to obtain a matching detection result of the predicted slot.

[0220] In one embodiment, the specific implementation manner for the slot recognition module 13 to detect the matching between the text data and the predicted slot in the language understanding model based on the slot detection hint information to obtain a matching detection result of the predicted slot includes:

[0221] In the language understanding model, determine the text slot value in the text data that matches the predicted slot as the verification slot value;

[0222] Determine the defined slot value list corresponding to the predicted slot as the target slot value list;

[0223] Perform a similarity analysis on the verification slot value and the target slot value list to obtain a similarity analysis result between the verification slot value and the target slot value list;

[0224] If the similarity analysis result indicates that there is a similarity between the verification slot value and the target slot value list, determine that the matching detection result of the predicted slot is a pass result;

[0225] If the similarity analysis result indicates that there is no similarity between the verification slot value and the target slot value list, determine that the matching detection result of the predicted slot is a fail result.

[0226] In one embodiment, the specific implementation manner in which the slot recognition module 13 performs a similarity analysis on the verification slot value and the target slot value list to obtain the similarity analysis result between the verification slot value and the target slot value list includes:

[0227] Calculate the word similarity between the verification slot value and each defined slot value in the target slot value list to obtain a set of word similarities;

[0228] Traverse the set of word similarities;

[0229] If there is a word similarity greater than the similarity threshold in the set of word similarities, it is determined that there is a defined slot value similar to the verification slot value in the target slot value list, and the similarity analysis result between the verification slot value and the target slot value list is determined as a similarity existence result;

[0230] If there is no word similarity greater than the similarity threshold in the set of word similarities, it is determined that there is no defined slot value similar to the verification slot value in the target slot value list, and the similarity analysis result between the verification slot value and the target slot value list is determined as a similarity non-existence result.

[0231] In one embodiment, the target slot value list contains the defined slot value S i , and the set of word similarities contains the word similarity between the verification slot value and the defined slot value S i ; i is a positive integer;

[0232] The specific implementation manner in which the slot recognition module 13 calculates the word similarity between the verification slot value and each defined slot value in the target slot value list to obtain a set of word similarities includes:

[0233] Obtain a first word vector for characterizing the verification slot value and a second word vector for characterizing the defined slot value S i ;

[0234] Calculate the vector distance between the first word vector and the second word vector;

[0235] Determine the word similarity between the verification slot value and the defined slot value S i according to the vector distance.

[0236] In one embodiment, the specific implementation manner in which the slot recognition module 13 determines the word similarity between the verification slot value and the defined slot value S i according to the vector distance includes:

[0237] Obtain a distance mapping table; the distance mapping table contains the mapping relationship between the configured distance interval set and the similarity set, and there is a mapping relationship between a configured distance interval in the configured distance set and a similarity in the similarity set;

[0238] Determine the configuration distance interval to which the vector distance in the set of configured distance intervals belongs as the target configuration distance interval;

[0239] Determine the similarity in the similarity set that has a mapping relationship with the target configuration distance interval as the word similarity between the verified slot value and the defined slot value S i therebetween.

[0240] In one embodiment, after the slot recognition module 13 obtains the text slots of the text data, the data processing device 1 further includes: an intention determination module 14 and a service module 15.

[0241] The intention determination module 14 is configured to determine the true intention indicated by the text data according to the text intention and the text slots;

[0242] The service module 15 is configured to provide services to the text providing object according to the true intention indicated by the text data; the text providing object refers to the object that provides the text data.

[0243] In one embodiment, the true intention is a navigation intention; the navigation intention refers to the intention of obtaining a navigation path for moving from a starting position point to a destination position point;

[0244] The specific implementation manner in which the service module 15 provides services to the text providing object according to the true intention indicated by the text data includes:

[0245] Generate a set of navigation paths for moving from the starting position point to the destination position point;

[0246] Perform a consumption duration analysis on each navigation path in the set of navigation paths to obtain the path consumption duration required for each navigation path;

[0247] Generate service indication information including the set of navigation paths and the path consumption duration required for each navigation path;

[0248] Push the service indication information to the text providing object.

[0249] Among them, for the specific implementation manners of the intention determination module 14 and the service module 15, reference may be made to the relevant descriptions in step S103 in the corresponding embodiment above, and details will not be elaborated here. Figure 2 which will not be elaborated here.

[0250] In the embodiments of the present application, defined slots are introduced in the text intention recognition of text data. By means of the defined slots corresponding to the intention, the text intention recognition can be divided into two independent atomic tasks (intention preliminary screening task and intention fine filtering task). Through task decomposition, the computational complexity of data reasoning can be reduced in each subdivided atomic task, thereby improving the recognition efficiency. At the same time, by introducing the information of this dimension of defined slots to perform fine filtering detection on the intention, the intentions that do not meet the requirements screened out initially can be filtered, and the text intention of the text data can be obtained more accurately. Similarly, in the text slot recognition of text data, defined slot values are introduced. By means of the defined slot values corresponding to the defined slots, the text slot recognition can be divided into two independent atomic tasks (slot preliminary screening task and slot fine filtering task). Through task decomposition, the computational complexity of data reasoning can be reduced in each subdivided atomic task, thereby improving the recognition efficiency. At the same time, by introducing the information of this dimension of defined slot values to perform fine filtering detection on the slot, the slots that do not meet the requirements screened out initially can be filtered, and the text slot of the text data can be obtained more accurately. When both the text intention and the text slot of the text data have high accuracy, the true intention determined according to the text intention and the text slot can also be more accurate. In addition, since the present application can directly use the defined slots to detect and filter the text intention, and use the defined slot values to detect and filter the text slot, even if the intention recognition solution provided by the present application is deployed in the model, there is no need to configure dialogue data to train the model. Only by configuring the intention, the slot, and the slot value, the model can be applied to the intention recognition scenario. Through the natural language understanding ability of the model, intention prediction and slot prediction can be performed, and then detection and filtering can be carried out through the defined slots and slot values, which can well reduce the data configuration threshold and improve the intention recognition efficiency and real-time performance.

[0251] Further, please refer to Figure 9 , Figure 9 which is a schematic structural diagram of a computer device provided by an embodiment of the present application. As Figure 9As shown in the figure, the above computer device 8000 may include: a processor 8001, a network interface 8004, and a memory 8005. In addition, the above computer device 8000 further includes: a user interface 8003 and at least one communication bus 8002. Among them, the communication bus 8002 is used to realize the connection and communication between these components. Among them, the user interface 8003 may include a display screen (Display) and a keyboard (Keyboard). Optionally, the user interface 8003 may further include a standard wired interface and a wireless interface. The network interface 8004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 8005 may be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. Optionally, the memory 8005 may further be at least one storage device located far from the aforementioned processor 8001. As Figure 9 shown, the memory 8005, as a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a device control application program.

[0252] In Figure 9 the computer device 8000 shown in the figure, the network interface 8004 can provide network communication functions; while the user interface 8003 is mainly used to provide an input interface for users; and the processor 8001 can be used to call the device control application program stored in the memory 8005 to achieve:

[0253] Obtain the text data to be recognized;

[0254] By defining an intent list and a set of defined slot lists, perform intent recognition on the text data to obtain the text intent of the text data; there is a one-to-one correspondence between the defined slot lists in the set of defined slot lists and the defined intents in the intent list;

[0255] By defining a set of defined slot value lists and the defined slot list corresponding to the text intent, perform slot recognition on the text data to obtain the text slots of the text data; there is a one-to-one correspondence between the defined slot value lists in the set of defined slot value lists and the defined slots in the defined slot list; the text intent and the text slots are used to reflect the services required by the text data.

[0256] It should be understood that the computer device 8000 described in the embodiments of the present application can execute the description of the data processing method in the corresponding embodiments described above Figures 2 to 6 and can also execute the description of the data processing device 1 in the corresponding embodiments described above Figure 8 and will not be elaborated here. In addition, the description of the beneficial effects of adopting the same method will not be elaborated either.

[0257] In addition, it should be noted here that: The embodiments of the present application also provide a computer-readable storage medium, and the computer-readable storage medium stores a computer program executed by the aforementioned computer device 8000 for data processing. The computer program includes program instructions. When the processor executes the program instructions, it can execute the description of the aforementioned data processing method in the corresponding embodiments mentioned above. Therefore, the description will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated either. For the technical details not disclosed in the embodiments of the computer-readable storage medium involved in the present application, please refer to the description of the method embodiments of the present application. Figures 2 to 6 The description of the corresponding embodiments of the above data processing method will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated either. For the technical details not disclosed in the embodiments of the computer-readable storage medium involved in the present application, please refer to the description of the method embodiments of the present application.

[0258] The above computer-readable storage medium may be the data processing device provided in any of the foregoing embodiments or an internal storage unit of the above computer device, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device. Further, the computer-readable storage medium may also include both the internal storage unit and the external storage device of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store the data that has been output or will be output.

[0259] In one aspect of the present application, a computer program product is provided. The computer program product includes a computer program, and the computer program is stored in a computer-readable storage medium. The processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the method provided in one aspect of the embodiments of the present application.

[0260] The terms "first", "second", etc. in the description, claims and drawings of the embodiments of the present application are used to distinguish different objects, rather than to describe a specific order. In addition, the term "comprising" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or equipment that includes a series of steps or units is not limited to the listed steps or modules, but optionally further includes steps or modules not listed, or optionally further includes other step units inherent to these processes, methods, devices, products or equipment.

[0261] In the embodiments of the present application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other relevant parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of the overall module or unit that includes the function of the module or unit.

[0262] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0263] The methods and related devices provided in the embodiments of the present application are described with reference to the method flowcharts and / or structural schematic diagrams provided in the embodiments of the present application. Specifically, each process and / or block of the method flowchart and / or structural schematic diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one process or multiple processes and / or structural schematic Figure 1 one block or multiple blocks. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device realizes the functions specified in Figure 1 one process or multiple processes and / or structural schematic Figure 1 one block or multiple blocks. These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in Figure 1 one process or multiple processes and / or structural schematic one block or multiple blocks.

[0264] The above disclosure is only for the preferred embodiments of the present application. Of course, the scope of rights of the present application cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.

Claims

1. A data processing method, characterized in that, Including: Obtain text data to be recognized; Perform intent recognition on the text data by defining an intent list and a set of defined slot lists to obtain the text intent of the text data; There is a one-to-one correspondence between the defined slot lists in the set of defined slot lists and the defined intents in the defined intent list; Perform slot recognition on the text data by defining a set of slot value lists and the defined slot list corresponding to the text intent to obtain the text slots of the text data; there is a one-to-one correspondence between the defined slot value lists in the set of defined slot value lists and the defined slots in the defined slot list; the text intent and the text slots are used to reflect the service required by the text data.

2. The method according to claim 1, wherein The performing intent recognition on the text data by defining an intent list and a set of defined slot lists to obtain the text intent of the text data includes: Perform intent semantic recognition on the text data by defining an intent list to obtain the predicted intent of the text data; the defined intent list includes the predicted intent; Obtain the defined slot list corresponding to the predicted intent from the set of defined slot lists; Perform intent adaptability detection on the predicted intent by the defined slot list corresponding to the predicted intent to obtain the adaptability detection result of the predicted intent; If it is determined that the adaptability detection result indicates that the predicted intent is adaptable to the text data, then determine the predicted intent as the text intent of the text data.

3. The method according to claim 2, wherein The performing intent semantic recognition on the text data by defining an intent list to obtain the predicted intent of the text data includes: Obtain historical text data that has a contextually coherent association with the text data; Generate intent recognition prompt information according to the defined intent list, the historical text data, and the text data, and input the intent recognition prompt information into a language understanding model; In the language understanding model, perform intent semantic recognition on the text data based on the intent recognition prompt information, and recognize the predicted intent of the text data from the defined intent list.

4. The method according to claim 2, characterized in that, The performing intent adaptability detection on the predicted intent by the defined slot list corresponding to the predicted intent to obtain the adaptability detection result of the predicted intent includes: Determine the defined slot list corresponding to the predicted intent as the first target slot list; Obtain the intent description information configured for the predicted intent, and historical text data that has a contextually coherent association with the text data; Generate intent detection prompt information according to the predicted intent, the intent description information of the predicted intent, the first target slot list, the historical text data, and the text data, and input the intent detection prompt information into a language understanding model; In the language understanding model, analyze the adaptability between the text data and the predicted intent based on the intent detection prompt information to obtain the adaptability detection result of the predicted intent.

5. The method according to claim 1, wherein Slot recognition of the text data is performed by defining a set of slot value lists corresponding to the text intent and a defined slot list, and the text slots of the text data are obtained, including: Determine the defined slot list corresponding to the text intent as the second target slot list; Perform slot matching recognition on the text data through the second target slot list to obtain the predicted slots of the text data; the second target slot list includes the predicted slots; Obtain the defined slot value list corresponding to the predicted slot from the set of defined slot value lists; Perform a matching detection on the predicted slot through the defined slot value list corresponding to the predicted slot to obtain a matching detection result of the predicted slot; If it is determined that the matching detection result indicates that the predicted slot matches the text data, then determine the predicted slot as the text slot of the text data.

6. The method according to claim 5, characterized in that, The step of performing slot matching recognition on the text data through the second target slot list to obtain the predicted slots of the text data includes: Obtain the intent description information configured for the text intent and the historical text data that has a contextually coherent association with the text data; Generate slot matching prompt information based on the text intent, the intent description information of the text intent, the second target slot list, the historical text data, and the text data, and input the slot matching prompt information into the language understanding model; In the language understanding model, perform slot matching recognition on the text data based on the slot matching prompt information to obtain a slot recognition result; the slot recognition result includes the predicted slot matched and recognized by the language understanding model from the second slot list, the text slot value in the text data that matches the predicted slot, and the reasoning description corresponding to the predicted slot; Determine the predicted slot included in the slot recognition result as the predicted slot of the text data.

7. The method according to claim 6, characterized in that, The step of performing a matching detection on the predicted slot through the defined slot value list corresponding to the predicted slot to obtain a matching detection result of the predicted slot includes: Obtain the historical text data that has a contextually coherent association with the text data; Generate slot detection prompt information based on the text intent, the predicted slot under the text intent, the defined slot value list corresponding to the predicted slot, the text slot value in the text data that matches the predicted slot, the reasoning description corresponding to the predicted slot, the historical text data, and the text data, and input the slot detection prompt information into the language understanding model; In the language understanding model, detect the matching between the text data and the predicted slot based on the slot detection prompt information to obtain a matching detection result of the predicted slot.

8. The method according to claim 7, characterized in that The step of detecting the matching between the text data and the predicted slot based on the slot detection prompt information in the language understanding model to obtain a matching detection result of the predicted slot includes: In the language understanding model, determine the text slot value in the text data that matches the predicted slot as the verification slot value; Determine the list of defined slot values corresponding to the predicted slot as the target slot value list; Perform a similarity analysis on the verification slot value and the target slot value list to obtain a similarity analysis result between the verification slot value and the target slot value list; If the similarity analysis result indicates that there is a similarity between the verification slot value and the target slot value list, determine that the matching detection result of the predicted slot is a pass result; If the similarity analysis result indicates that there is no similarity between the verification slot value and the target slot value list, determine that the matching detection result of the predicted slot is a fail result.

9. The method according to claim 8, characterized in that, The performing a similarity analysis on the verification slot value and the target slot value list to obtain a similarity analysis result between the verification slot value and the target slot value list includes: Calculate the word similarity between the verification slot value and each defined slot value in the target slot value list to obtain a set of word similarities; Traverse the set of word similarities; If there is a word similarity greater than the similarity threshold in the set of word similarities, determine that there is a defined slot value in the target slot value list that is similar to the verification slot value, and determine the similarity analysis result between the verification slot value and the target slot value list as a similarity existence result; If there is no word similarity greater than the similarity threshold in the set of word similarities, determine that there is no defined slot value in the target slot value list that is similar to the verification slot value, and determine the similarity analysis result between the verification slot value and the target slot value list as a similarity non - existence result.

10. The method according to claim 9, wherein The target slot value list contains the defined slot value S i , and the word similarity set contains the word similarity between the verification slot value and the defined slot value S i ; i is a positive integer; The calculating the word similarity between the verification slot value and each defined slot value in the target slot value list to obtain a set of word similarities includes: Obtain a first word vector for characterizing the verification slot value and a second word vector for characterizing the defined slot value S i ; Calculate the vector distance between the first word vector and the second word vector; Determine the word similarity between the verification slot value and the defined slot value S according to the vector distance i therebetween.

11. The method according to claim 1, characterized in that, After obtaining the text slots of the text data, the method further includes: Determine the true intention indicated by the text data according to the text intention and the text slots; Provide services to the text providing object according to the true intention indicated by the text data; the text providing object refers to the object that provides the text data.

12. The method according to claim 11, wherein The true intention is a navigation intention; the navigation intention refers to the intention of obtaining a navigation path from a starting position point to a destination position point; The providing services to the text providing object according to the true intention indicated by the text data includes: Generate a set of navigation paths from the starting position point to the destination position point; Perform a consumption duration analysis on each navigation path in the set of navigation paths to obtain the path consumption duration required for each navigation path; Generate service indication information including the set of navigation paths and the path consumption duration required for each navigation path; Push the service indication information to the text providing object.

13. A data processing device, characterized in that, including: A data acquisition module for acquiring text data to be recognized; An intent recognition module, configured to perform intent recognition on the text data by defining an intent list and a set of defined slot lists, so as to obtain the text intent of the text data; there is a one-to-one correspondence between the defined slot lists in the set of defined slot lists and the defined intents in the defined intent list; A slot recognition module, configured to perform slot recognition on the text data by defining a set of defined slot value lists and the defined slot list corresponding to the text intent, so as to obtain the text slots of the text data; there is a one-to-one correspondence between the defined slot value lists in the set of defined slot value lists and the defined slots in the defined slot list; the text intent and the text slots are used to reflect the service required by the text data.

14. A computer device, characterized in that, Comprising: A processor, a memory, and a network interface; The processor is connected to the memory and the network interface, wherein the network interface is used to provide network communication functions, the memory is used to store computer programs, and the processor is used to call the computer programs so that the computer device executes the method according to any one of claims 1-12.

15. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, and the computer program is suitable for being loaded and executed by the processor to execute the method according to any one of claims 1-12.

16. A computer program product, characterized in that, The computer program product includes a computer program, the computer program is stored in a computer-readable storage medium, and the computer program is suitable for being read and executed by the processor so that a computer device having the processor executes the method according to any one of claims 1-12.

Images