Information recommendation method and device, electronic equipment and storage medium

Abstract

The application discloses an information recommendation method and device, an electronic device, and a storage medium. The information recommendation method comprises the following steps: obtaining dialogue data in a dialogue scene; determining user intention information in the dialogue scene by using an intention recognition model and based on the dialogue data; determining a recommendation question matched with the user intention information from preconfigured recommendation questions as a target question, wherein the recommendation question is generated by a large language model in advance according to different intention information output by the intention recognition model; and recommending the target question. The method can accurately recommend relevant questions to a user in a dialogue scene.

Description

Technical Field

[0001] This application relates to the field of electronic equipment technology, and more specifically, to an information recommendation method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the rapid advancement of technology and living standards, electronic devices (such as smartphones and tablets) have become commonplace in people's lives. Most current electronic devices are equipped with intelligent assistants to assist users in using them, allowing for daily interactions. In these interactions, intelligent assistants typically provide recommended questions for users to choose from, thereby improving efficiency. However, the accuracy of these recommended questions in conversational scenarios remains insufficient. Summary of the Invention

[0003] This application proposes an information recommendation method, apparatus, electronic device, and storage medium that can improve the accuracy of recommending relevant questions to users in dialogue scenarios.

[0004] In a first aspect, embodiments of this application provide an information recommendation method, the method comprising: acquiring dialogue data in a dialogue scenario; using an intent recognition model and based on the dialogue data, determining user intent information in the dialogue scenario; determining a recommendation question matching the user intent information from a pre-configured set of recommendation questions as a target question, wherein the recommendation question is pre-generated using a large language model and based on different intent information output by the intent recognition model; and recommending the target question.

[0005] Secondly, embodiments of this application provide an information recommendation device, the device comprising: a data acquisition module, an intent recognition module, a question determination module, and a question recommendation module, wherein the data acquisition module is used to acquire dialogue data in a dialogue scenario; the intent recognition module is used to determine user intent information in the dialogue scenario using an intent recognition model and based on the dialogue data; the question determination module is used to determine a recommended question matching the user intent information from a pre-configured list of recommended questions as a target question, the recommended question being pre-generated using a large language model and based on different intent information output by the intent recognition model; and the question recommendation module is used to recommend the target question.

[0006] Thirdly, embodiments of this application provide an electronic device, including: one or more processors; a memory; and one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to perform the information recommendation method provided in the first aspect above.

[0007] Fourthly, embodiments of this application provide a computer-readable storage medium storing program code, which can be invoked by a processor to execute the information recommendation method provided in the first aspect above.

[0008] The solution provided in this application acquires dialogue data from a dialogue scenario, utilizes an intent recognition model, and determines user intent information based on the dialogue data. It then selects a target question from a pre-configured pool of target questions that matches the user intent information. This target question is generated in advance using a large language model and based on different intent information output by the intent recognition model. Because the target question is generated using a large language model and the recognized intent information, the determined target question better matches the user's intent, resulting in more accurate recommendations. Furthermore, since the target question is pre-configured, there is no need to re-invoke the large language model to generate target questions during the dialogue, thereby improving the response speed for recommending relevant questions to the user. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic diagram of the application environment provided in the embodiments of this application is shown.

[0011] Figure 2 A flowchart illustrating an information recommendation method according to an embodiment of this application is shown.

[0012] Figure 3 A schematic diagram of an interface provided in an embodiment of this application is shown.

[0013] Figure 4 A flowchart illustrating an information recommendation method according to another embodiment of this application is shown.

[0014] Figure 5 A flowchart illustrating an information recommendation method according to yet another embodiment of this application is shown.

[0015] Figure 6 A flowchart illustrating an information recommendation method according to another embodiment of this application is shown.

[0016] Figure 7A flowchart illustrating an information recommendation method according to yet another embodiment of this application is shown.

[0017] Figure 8 A block diagram of an information recommendation device according to an embodiment of this application is shown.

[0018] Figure 9 This is a block diagram of an electronic device for performing an information recommendation method according to an embodiment of this application.

[0019] Figure 10 This is a storage unit in this application embodiment for storing or carrying program code that implements the information recommendation method according to this application embodiment. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0021] With the rapid development of artificial intelligence and mobile communication technologies, smart assistants have become an indispensable part of electronic devices. When using a smart assistant, users typically engage in conversation using voice, text, and images. During the conversation, the smart assistant can usually provide answers based on the user's input. Furthermore, it can recommend questions for the user to choose from, and after the user selects a question, the smart assistant can continue to provide services based on that selection.

[0022] In related technologies, intelligent assistants typically rely on preset command sets, rule matching, and simple keyword matching to identify user intent and provide corresponding recommended questions when offering recommendations. These methods work reasonably well for simple, direct commands, but they often struggle to accurately output recommended questions that match the user's intent when faced with complex, fuzzy, or multimodal user queries. This results in poor recommendation performance and an inability to meet user needs.

[0023] To address the aforementioned problems, the inventors have proposed an information recommendation method, apparatus, electronic device, and storage medium as described in the embodiments of this application. These methods enable the output recommendation questions to better align with user intent, resulting in more accurate recommendations. Furthermore, since the recommendation questions are pre-configured, there is no need to re-invoke the large language model to generate recommendation questions during the dialogue process, thereby improving the response speed for recommending relevant questions to the user. The specific information recommendation method will be described in detail in subsequent embodiments.

[0024] The scenarios involved in the embodiments of this application will be introduced below.

[0025] like Figure 1 As shown, in Figure 1 The scenario shown includes an electronic device 100 and a server 200. The electronic device 100 can be any electronic device with communication and storage functions, including but not limited to a PC (Personal Computer), PDA (Tablet PC), smart TV, smartphone, smart wearable device, or other smart communication device with network connectivity. The server 200 can be a single server (network access server), a server cluster consisting of several servers (cloud server), or a cloud computing center (database server).

[0026] Electronic device 100 and server 200 can interact via a network to receive or send information. In this embodiment, server 200 may be equipped with a large language model and an intent recognition model. After server 200 generates recommendation questions that match various user intent information based on the large language model and different intent information output by intent recognition model, it can send the generated recommendation questions to electronic device 100 to pre-configure the recommendation questions in electronic device 100. When electronic device 100 needs to recommend questions to users, it can acquire dialogue data in the dialogue scenario, use the intent recognition model, and determine the user intent information in the dialogue scenario based on the dialogue data. It can then determine the recommendation question that matches the user intent information from the pre-configured recommendation questions as the target question and recommend the target question.

[0027] The information recommendation method provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0028] Please see Figure 2 , Figure 2 A flowchart illustrating an embodiment of the information recommendation method provided in this application is shown. In a specific embodiment, the information recommendation method is applied to, for example... Figure 8 The information recommendation device 600 shown and the electronic device 100 configured with the information recommendation device 600 are also shown. Figure 9 The following will use an electronic device as an example to illustrate the specific process of this embodiment. Of course, it is understood that the electronic device used in this embodiment can be a smartphone, tablet computer, smartwatch, e-reader, etc., and is not limited thereto. The following will focus on... Figure 2 The process shown will be described in detail. The information recommendation method may specifically include the following steps:

[0029] Step S110: Obtain dialogue data in the dialogue scenario.

[0030] In this embodiment, the electronic device can acquire dialogue data within a dialogue scenario to determine the user's intent within that scenario. Subsequently, based on the identified user intent, the device can recommend questions for the user to choose from within the dialogue scenario. The dialogue scenario can be any type of dialogue scenario, and the dialogue data can include the dialogue content within that scenario.

[0031] In some implementations, the above dialogue scenario can be a dialogue with a smart assistant. Electronic devices may have a smart assistant installed. The smart assistant can be an application based on artificial intelligence technology. Smart assistants are typically used to provide users with various services and assistance. They usually possess capabilities such as speech recognition, natural language processing, and machine learning, enabling them to converse with users, understand their needs, and then provide corresponding information, suggestions, or perform tasks.

[0032] In one possible implementation, the electronic device may display a dialogue interface with a smart assistant. This dialogue interface can be understood as a chat interface with a virtual conversation object corresponding to the smart assistant. The virtual conversation object can be a virtual object constructed by the electronic device for interaction with the user; when in a chat context, this virtual object can be understood as a virtual conversation object. The dialogue data may include text, images, videos, etc., entered by the user in the chat interface, as well as the responses from the virtual conversation object.

[0033] Step S120: Using an intent recognition model and based on the dialogue data, determine the user intent information in the dialogue scenario.

[0034] In this embodiment, after acquiring the above dialogue data, user intent information in the above dialogue scenario can be identified based on the acquired dialogue data. Specifically, the electronic device can identify user intent information based on an intent recognition model. The intent recognition model is a key component of Natural Language Processing (NLP), used to understand and analyze user intent. Intent recognition models can be Support Vector Machines (SVM), Decision Tree models, Convolutional Neural Networks (CNN), Transformers (self-attention mechanism) models, BERT models, etc., and the specific model type is not limited. The above intent recognition model can be deployed in an electronic device or on a server. When deployed on a server, the user intent information can be obtained through server recognition.

[0035] In some implementations, when using the above intent recognition model and determining user intent information in the dialogue scenario based on the acquired dialogue data, the above dialogue data can be converted into corresponding text information; then the text information is input into the intent recognition model to obtain the output of the above intent recognition model, which serves as the user intent information in the above dialogue scenario.

[0036] In one possible implementation, the text information obtained after the above transformation can be segmented into multiple keywords; then, these keywords can be quantized into windowed word vectors; the obtained word vectors can be input into an intent recognition model to obtain the probabilities of various intents output by the intent recognition model; finally, based on the probabilities of various intents, the intents with probabilities greater than a specified probability can be determined as the intents to be recognized. The intent recognition model can be trained on an initial model based on the word vectors of keywords corresponding to a large number of sentences and the labeled intents. The specific type of the initial model is not limited; for example, it can be a convolutional neural network.

[0037] Step S130: Determine a recommendation question that matches the user intent information from the pre-configured recommendation questions as the target question. The recommendation question is generated in advance through a large language model and based on different intent information output by the intent recognition model.

[0038] In this embodiment, after identifying the user intent information in the above dialogue scenario, a recommendation question matching the user intent information can be determined from the pre-configured recommendation questions, which will then be used as the target question to be recommended to the user. The pre-configured recommendation questions are generated using a Large Language Model (LLM) based on different intent information output by the intent recognition model. In other words, the target question determined above is also generated using the LLM. Because the LLM has strong understanding and generation capabilities, the recommendation question determined from the user intent information is more consistent with the user intent, thus improving the accuracy of the determined recommendation question.

[0039] Furthermore, although the final target question is generated by the large language model, it is not necessary to call the large language model in the process of determining the recommendation question to recommend to the user. Therefore, it can improve the efficiency of determining the recommendation question to recommend to the user, thereby improving the response speed of recommending relevant questions to the user.

[0040] In some implementations, the electronic device may store a mapping relationship between different user intent information and recommendation questions. After each user intent information in a dialogue scenario is identified, the recommendation question corresponding to the identified user intent information can be determined based on the stored mapping relationship.

[0041] In some implementations, the user intent model described above can output results for various intent categories. For each intent category, the structured output capability of the large language model can be utilized to output corresponding guidance and recommendations. The large language model is a model based on deep learning and natural language processing techniques. It is trained using a large amount of text data to learn language understanding and generation capabilities. The large language model can handle complex natural language tasks such as text classification, question answering, and dialogue.

[0042] In some implementations, after determining the recommended questions that match the user's intent information from a pre-configured pool of recommended questions, further filtering can be performed from the determined recommended questions based on dialogue data within the dialogue scenario to select target questions for recommendation to the user. For example, if the user's intent information is "food search," the dialogue data includes "the intelligent assistant's response: This is food A," and the user's input image was obtained through taking a picture and selecting an image, then combining this information, it can be determined that the user's environment is likely eating, and their next intent may be to post content on a social media platform. Therefore, from the recommended questions matching "food search," the recommended question "create a caption for a social media platform" can be determined as the target question for recommendation to the user.

[0043] Step S140: Make recommendations for the target problem.

[0044] In this embodiment of the application, after determining the above target questions, recommendations can be made regarding these target questions. Specifically, the electronic device can output the above target questions in the above dialogue scenario for the user to choose from.

[0045] In some implementations, the electronic device can display the target question in a dialogue interface within the above-described dialogue scenario; for example, the target question can be displayed in the dialogue interface of a smart assistant. After displaying the target question in the above dialogue interface, operations within the dialogue interface can be detected; if a selection operation for the target question is detected, a query for the target question can be performed in response to the selection operation to output an answer to the target question.

[0046] In one possible implementation, the electronic device can utilize a large language model to generate an answer to the target question. For example, the electronic device can send the target question to a server deployed with a large language model. After receiving the target question, the server can input the target question into the large language model and obtain the generated content output by the large language model based on the target question. This generated content can then serve as the answer to the target question, and the server returns the answer to the electronic device. Correspondingly, the electronic device can receive the answer returned by the server.

[0047] For example, please refer to Figure 3 In the dialogue interface A1 with the intelligent assistant, after the user inputs an image, the intelligent assistant can respond with the image recognition result based on the image. Furthermore, based on the dialogue data in the dialogue scenario, after recognizing the user's intent information, it can determine a recommended question matching the user's intent information from a pre-configured list of recommended questions and display the determined recommended question on the dialogue interface, such as... Figure 3 The dialogue interface A1 shown includes the questions "What are the nutritional components of this food?" and "Pair it with a social media post".

[0048] In some implementations, the electronic device can also broadcast the target question via voice in the above dialogue scenario. After broadcasting the target question, the electronic device can detect the user's voice input; then it can recognize the detected voice, and if a selection instruction for the target question is recognized, it can respond to the selection instruction to query the target question in order to output the answer to the target question.

[0049] The information recommendation method provided in this application identifies user intent information in a dialogue scenario based on the dialogue data, and then determines the recommendation questions to be recommended to the user from a pre-configured set of recommendation questions. The pre-configured recommendation questions are generated using a large language model and the identified intent information, thus the determined recommendation questions are more accurate and better match the user's intent, resulting in more accurate recommendations to the user. In addition, since the recommendation questions are pre-configured, there is no need to call the large language model again to generate recommendation questions during the dialogue process, thereby improving the response speed of recommending relevant questions to the user.

[0050] Please see Figure 4 , Figure 4 A flowchart illustrating another embodiment of the information recommendation method provided in this application is shown. This information recommendation method is applied to the aforementioned electronic device, and will be discussed below. Figure 4 The process shown will be described in detail. The information recommendation method may specifically include the following steps:

[0051] Step S210: Obtain dialogue data in the dialogue scenario.

[0052] In this embodiment, step S210 can be referred to the content of other embodiments, and will not be repeated here.

[0053] Step S220: Using an intent recognition model and based on the dialogue data, determine the user intent information in the dialogue scenario, wherein the user intent information includes intent categories.

[0054] In this application embodiment, the user intent information in the above dialogue scenario identified based on the above dialogue data may include intent categories. Intent categories can represent the category to which the user intent belongs, such as categories like food search, news search, music search, location search, mobile phone function operation, sensitive topics, etc.

[0055] In some implementations, the above intent categories can be obtained by clustering different intent information. The method for clustering different intent information can utilize machine learning methods for classification, such as K-means clustering, DBSCAN clustering, etc., and the specific clustering method is not limited.

[0056] Step S230: Determine at least one recommendation question that matches the intent category from the pre-configured recommendation questions as the target question. The recommendation question is generated in advance by a large language model and based on different intent information output by the intent recognition model.

[0057] In this embodiment of the application, after obtaining the above intent categories, at least one recommended question that matches the above intent categories can be queried from the pre-configured recommended questions, thereby obtaining the target question to be recommended to the user.

[0058] In some implementations, after retrieving at least one recommended question matching the above intent category from a pre-configured list of recommended questions, there may be a large number of pre-configured recommended questions corresponding to the above intent categories. Therefore, when recommending these questions to the user, there may be many questions that the user will not select. Thus, the electronic device can further determine the number of recommended questions matching the intent category and compare this number with a first threshold. Based on the comparison result, if the number is less than the first threshold, then the recommended questions matching the intent category can be directly identified as the target questions. The first threshold can be 2, 3, 5, etc., and its specific value is not limited.

[0059] Furthermore, after comparing the number of recommended questions matching the above intent categories with a first threshold, if the number is greater than or equal to the first threshold, it indicates that a large number of recommended questions have been identified, thus allowing for further filtering. In this case, the electronic device can determine the intent score for each recommended question matching the intent category; then, it can identify recommended questions whose intent scores meet the target score criteria as target questions for recommendation to the user. The intent score for each recommended question can be determined based on historical feedback data and popularity data for that question. Understandably, the higher the intent score, the greater the probability that the recommended question will be selected by the user; therefore, it is possible to identify recommended questions whose intent scores meet the target score criteria, further improving the accuracy of the identified recommended questions. The target score criteria may include the highest intent score or an intent score greater than a preset score; the specific preset score criteria are not limited.

[0060] In some implementations, the historical feedback data may include the number of times the recommended question was selected in the past, and the popularity data may include the number of times multiple users queried the recommended question in the past, as counted by the server. This server may be the server used to query the recommended question, i.e., the server that provides the answers to the recommended question. When determining the intent score, a first score corresponding to the recommended question can be determined based on the number of historical selections; a second score can be determined based on the number of historical queries; then, the intent score is determined based on the first and second scores. This intent score is positively correlated with the first and second scores; that is, if the second scores are the same, the higher the first score, the higher the intent score; and if the first scores are the same, the higher the second score, the higher the intent score. Understandably, since the number of historical selections reflects users' personal habits and preferences, and the number of historical queries reflects the popularity of the recommended question, determining the intent score based on the number of historical selections and queries, and selecting recommended questions based on the intent score, can be more accurate.

[0061] In one possible implementation, when determining the first score based on the historical selection count of the recommendation question, the sum of the historical selection counts of all recommendation questions matching the intent category can be obtained as the first sum. Then, for each recommendation question matching the intent category, the ratio of the historical selection count of each recommendation question to the first sum is obtained to obtain the normalized value corresponding to each recommendation question, which is used as the first score corresponding to each recommendation question.

[0062] In one possible implementation, when determining the second score based on the historical query counts of the recommendation question, the sum of the historical query counts of all recommendation questions matching the intent category can be obtained as the second sum. Then, for each recommendation question matching the intent category, the ratio of the historical query count of each recommendation question to the second sum is obtained to obtain the normalized value corresponding to each recommendation question, which is used as the second score corresponding to each recommendation question.

[0063] In one possible implementation, after each output of a recommended question, the electronic device can obtain feedback data based on the user's actions regarding the recommended question, and record this feedback data, thereby recording historical feedback data for each recommended question. Specifically, if the user selected one of the output recommended questions through an action, the selection count for that recommended question can be increased by one when recording the feedback data.

[0064] In one possible implementation, after receiving query requests from each user's electronic device for a recommendation question, the server can record the number of queries for each recommendation question, thereby recording the historical number of queries for each recommendation question.

[0065] Step S240: Make recommendations for the target problem.

[0066] In this embodiment, step S240 can be referred to the content of the foregoing embodiments, and will not be repeated here.

[0067] The information recommendation method provided in this application identifies the intent category in the dialogue scenario based on the dialogue data, determines the recommendation question matching the intent category from a pre-configured set of recommendation questions, and recommends the determined recommendation question to the user. The pre-configured recommendation question is generated by a large language model and the identified intent information, so the determined recommendation question is more accurate and better matches the user's intent, thus making the recommended question to the user more accurate. Furthermore, since the recommendation question is pre-configured, there is no need to call the large language model again to generate recommendation questions during the dialogue, thereby improving the response speed of recommending relevant questions to the user.

[0068] Please see Figure 5 , Figure 5 A flowchart illustrating another embodiment of the information recommendation method provided in this application is shown. This information recommendation method is applied to the aforementioned electronic device, and will be discussed below. Figure 5 The process shown will be described in detail. The information recommendation method may specifically include the following steps:

[0069] Step S310: Obtain sample dialogue data from different historical dialogue scenarios.

[0070] Unlike other embodiments, this embodiment also provides a method for generating pre-configured recommendation questions in various embodiments of this application. It is worth noting that the process of generating pre-configured recommendation questions can be performed in advance. Whenever the electronic device needs to determine the recommendation questions to recommend to the user, it can directly determine the corresponding recommendation questions from the pre-configured list, without having to generate pre-configured recommendation questions every time it needs to determine the recommendation questions to recommend to the user.

[0071] In this embodiment of the application, when generating recommendation questions with different user intent information, sample dialogue data from different historical dialogue scenarios can be obtained. These historical dialogue scenarios can cover different user intents. For example, user intent information can include multiple intent categories, so sample dialogue data from historical dialogue scenarios in which these intent categories appear can be obtained for these intent categories.

[0072] Step S320: Using the intent recognition model and based on the sample dialogue data, determine the user intent information in the historical dialogue scenario as sample intent information.

[0073] In this embodiment, after obtaining the above sample dialogue data, user intent information in the above historical dialogue scenarios is identified based on the obtained sample dialogue data, and used as sample intent information. The identified sample intent information can include various categories of intent information, so that after generating recommendation questions that match the sample intent information, the resulting recommendation questions can cover all possible user intents.

[0074] In some implementations, when using the above intent recognition model and based on the acquired historical dialogue data to determine the sample intent information in the sample dialogue scenario, the above historical dialogue data can be converted into corresponding text information; then the text information is input into the intent recognition model to obtain the output of the above intent recognition model, which serves as the sample intent information in the above sample dialogue scenario.

[0075] Step S330: Using a large language model and based on the sample intent information, generate a recommendation question that matches the sample intent information.

[0076] In this embodiment of the application, after obtaining various possible sample intent information, a recommendation question matching the sample intent information can be generated through a large language model based on the sample intent information.

[0077] In some implementations, the above sample intent information and prompts can be input into a large language model to obtain the recommendation question output by the large language model. The prompt refers to the input text or instruction provided by the model, guiding the large language model to generate a specific type of response. The prompt converts natural language text into machine-readable intent and embedding vectors, enabling the large model to understand and execute human instructions. Specifically, the prompt can be a question, a description, a task description, or even a portion of the dialogue history, and its function is to guide the model to generate an expected response or complete a specific task. In this implementation, the prompt is used to prompt the large language model to generate a recommendation question for the user based on the input sample intent information (i.e., generate a recommendation question). For example, it could be "Generate a query recommended to the user based on this intent information." For instance, if the identified sample intent information is "query running shoes," the large language model can generate recommendation questions such as "running shoe recommendations," "best running shoe brands," and "running shoe reviews" based on the prompt.

[0078] In one possible implementation, the sample intent information may include intent categories, such as food search, news search, music search, location search, mobile phone function operation, sensitive topics, etc. For each intent category, a corresponding recommendation question can be generated using a large language model and configured into the electronic device. Thus, when determining the question to recommend to the user, the electronic device can query the recommendation question that matches the identified intent category based on the identified intent category in the dialogue scenario.

[0079] Step S340: Obtain dialogue data in the dialogue scenario.

[0080] Step S350: Using the intent recognition model and based on the dialogue data, determine the user intent information in the dialogue scenario.

[0081] Step S360: Determine a recommendation question that matches the user intent information from the pre-configured recommendation questions as the target question. The recommendation question is generated in advance through a large language model and based on different intent information output by the intent recognition model.

[0082] Step S370: Make recommendations for the target problem.

[0083] In the embodiments of this application, steps S340 to S370 can be referred to the contents of other embodiments, and will not be repeated here.

[0084] The information recommendation method provided in this application embodiment also provides a method for generating pre-configured recommendation questions. The pre-configured recommendation questions are generated through a large language model and the identified intent information. Therefore, the generated recommendation questions can be more accurate. Thus, in the application process, the questions determined to recommend to the user based on the identified user intent information can be more accurate, thereby improving the accuracy of question recommendation.

[0085] Please see Figure 6 , Figure 6 A flowchart illustrating an information recommendation method provided in another embodiment of this application is shown. This information recommendation method is applied to the aforementioned electronic device, and will be discussed below. Figure 6 The process shown will be described in detail. The information recommendation method may specifically include the following steps:

[0086] Step S410: Obtain dialogue data in the dialogue scenario.

[0087] Step S420: Using the intent recognition model and based on the dialogue data, determine the user intent information in the dialogue scenario.

[0088] Step S430: Determine a recommendation question that matches the user intent information from the pre-configured recommendation questions as the target question. The recommendation question is generated in advance through a large language model and based on different intent information output by the intent recognition model.

[0089] Step S440: Make recommendations for the target problem.

[0090] In the embodiments of this application, steps S410 to S440 can be referred to the contents of other embodiments, and will not be repeated here.

[0091] Step S450: Obtain feedback data for the target problem and update the obtained feedback data to the historical feedback data corresponding to the target problem.

[0092] In this embodiment, after recommending the target questions identified above, the electronic device can obtain feedback data for the recommended target questions and update the historical feedback data corresponding to the target questions. The feedback data may include information on whether the user selected a target question, and the historical feedback data may include the historical number of times the target question was selected.

[0093] In one possible implementation, after each output of a recommended question, the electronic device can obtain feedback data based on the user's actions regarding the recommended question, and record this feedback data, thereby recording historical feedback data for each recommended question. Specifically, if the user selected one of the output recommended questions through an action, the selection count for that recommended question can be increased by one when recording the feedback data.

[0094] Step S460: If the historical feedback data meets the target feedback conditions, then the recommendation question matching the user intent information is updated.

[0095] In this embodiment of the application, based on the above historical feedback data, it can be determined whether the above historical feedback data meets the target feedback condition. The target feedback condition is used to determine whether to update the recommendation question that matches the above user intent information. If the historical feedback data meets the target feedback condition, it means that the recommendation effect of the above target question is not good. Therefore, the recommendation question that matches the above user intent information in the pre-configured recommendation question can be updated to improve the accuracy of question recommendation when the same user intent information is identified again in the future, thereby improving the recommendation effect of question recommendation.

[0096] In some implementations, the above target feedback condition may include the ratio of the historical number of selections for the above target problem to the historical number of recommendations for the above target problem being less than a target ratio.

[0097] In some implementations, if the historical feedback data of the target question meets the target feedback condition, the number of recommended questions matching the user intent information can be further determined, and this number can be compared with a second threshold. Based on the comparison result, if the number of questions is greater than or equal to the second threshold, it indicates that there are many recommended questions matching the user intent information, and therefore, recommended questions that meet the target feedback condition can be deleted from the list of recommended questions matching the user intent information. In other words, for each recommended question matching the user intent information, it can be determined individually whether the historical feedback data of each recommended question meets the target feedback condition. If the historical feedback data of any recommended question does not meet the target feedback condition, that recommended question can be deleted from the list of recommended questions matching the user intent information. The specific number of the second threshold is not limited; for example, it can be 3, 5, 7, etc.

[0098] Furthermore, after comparing the number of recommended questions matching the user's intent information with a second threshold, if the number of questions is less than the second threshold, a new recommended question can be generated using a large language model based on the dialogue data. This new recommended question then matches the user's intent information. Understandably, if the number of questions is less than the second threshold, it means there are fewer recommended questions matching the user's intent information. Therefore, deleting some recommended questions cannot improve the recommendation effect. Thus, a large language model can be used to regenerate recommended questions, updating the generated questions to match the user's intent information. Moreover, when regenerating recommended questions, the dialogue data is directly input into the large language model. The large language model understands the user's intent and generates recommended questions, thereby improving the accuracy of the generated recommended questions.

[0099] The information recommendation method provided in this application identifies user intent information in a dialogue scenario based on dialogue data, and then determines recommended questions to be recommended to the user from a pre-configured set of recommended questions. These pre-configured recommended questions are generated using a large language model and the identified intent information, resulting in more accurate and relevant recommendations that better align with user intent. Furthermore, since the recommended questions are pre-configured, there is no need to re-invoke the large language model during the dialogue, thus improving the response speed for recommending relevant questions to the user. Additionally, after recommending a question to the user, the configured recommended questions are updated based on user feedback data, ensuring the accuracy of the pre-configured recommended questions in the electronic device, further improving the accuracy and effectiveness of the question recommendation.

[0100] Please see Figure 7 , Figure 7 A flowchart illustrating an information recommendation method according to another embodiment of this application is shown. This information recommendation method is applied to the aforementioned electronic device, and will be discussed below. Figure 7 The process shown will be described in detail. The information recommendation method may specifically include the following steps:

[0101] Step S510: Obtain dialogue data in the dialogue scenario, wherein the dialogue data includes multiple types of data.

[0102] In this embodiment, the dialogue data can include various types of data, and the dialogue data can be multimodal data. These various types of data can include text, voice, images, documents, video, etc. Using multimodal data to identify user intent ensures comprehensive recognition of user intent and improves the accuracy of intent recognition.

[0103] Step S520: Based on the information extraction model, extract the input information corresponding to each type of data from the various types of data.

[0104] In this embodiment, after acquiring the various data types mentioned above, input information corresponding to each data type can be extracted from the data based on an information extraction model. Each data type can correspond to a separate information extraction model. For example, an image recognition model can be used to identify objects, scenes, and text in an image, while a document analysis model can be used to extract text content and structured data from a document. Through this method, various data types can be converted into information that can be input into the intent recognition model, enabling the intent recognition model to accurately identify user intent by combining the multiple data types.

[0105] Step S530: Input the input information corresponding to each type of data into the intent recognition model to obtain the intent recognition result output by the intent recognition model, which serves as the user intent information in the dialogue scenario.

[0106] In this embodiment of the application, after obtaining the input information corresponding to each type of data, the input information corresponding to each type of data can be input into the intent recognition model to obtain the intent recognition result output by the intent recognition model, and the obtained intent recognition result is used as the user intent information in the above dialogue scenario.

[0107] In some implementations, the BERT model is used for intent recognition due to its advantages in handling complex language structures and understanding contextual information. The BERT model has powerful transfer learning capabilities; through fine-tuning, it can easily adapt to various specific NLP tasks, such as text classification, named entity recognition, question answering, and sentiment analysis. In this implementation, by fine-tuning the BERT model, it can be adapted to specific application scenarios and user groups, providing more accurate intent recognition results. When inputting the input information corresponding to each of the above data types into the BERT model, the input information corresponding to multiple data types can be combined and input into the BERT model to obtain the intent recognition result.

[0108] Optionally, the output of the BERT model can be an intent category. For example, the input format of the BERT model can be: <image description information>, <image classification result>, <answer result>, <user's question>, and the output of the BERT model can be an intent category such as food search, news search, music search, location search, mobile phone function operation, or sensitive topic.

[0109] Step S540: Determine a recommendation question that matches the user intent information from the pre-configured recommendation questions as the target question. The recommendation question is generated in advance through a large language model and based on different intent information output by the intent recognition model.

[0110] Step S550: Make recommendations for the target problem.

[0111] In the embodiments of this application, steps S540 and S550 can be referred to the contents of other embodiments, and will not be repeated here.

[0112] The information recommendation method provided in this application identifies user intent information in a dialogue scenario based on dialogue data, and then determines recommended questions to be recommended to the user from a pre-configured set of recommendation questions. These pre-configured recommendation questions are generated using a large language model and the identified intent information, resulting in more accurate and relevant recommendations that better align with user intent. Furthermore, since the recommendation questions are pre-configured, there is no need to re-invoke the large language model during the dialogue process, thus improving the response speed for recommending relevant questions to the user. Additionally, when identifying user intent information in a dialogue scenario using the intent recognition model, multimodal information from the dialogue scenario is incorporated to identify user intent information. This allows for in-depth analysis of the user's query and contextual information within the dialogue scenario, enabling accurate identification of user intent information and further improving the accuracy of question recommendations.

[0113] Please see Figure 8 This diagram illustrates a structural block diagram of an information recommendation device 600 provided in an embodiment of this application. The information recommendation device 600 utilizes the aforementioned electronic device and includes: a data acquisition module 610, an intent recognition module 620, a question determination module 630, and a question recommendation module 640. Specifically, the data acquisition module 610 acquires dialogue data in a dialogue scenario; the intent recognition module 620 uses an intent recognition model and, based on the dialogue data, determines user intent information in the dialogue scenario; the question determination module 630 determines a target question from a pre-configured list of recommended questions that matches the user intent information; the target question is generated in advance using a large language model and based on different intent information output by the intent recognition model; and the question recommendation module 640 recommends the target question.

[0114] In some implementations, the user intent information includes intent categories, and the question determination module 630 may be specifically used to: determine at least one recommended question that matches the intent category from pre-configured recommended questions as the target question.

[0115] In one possible implementation, the problem determination module 630 may be specifically configured to: determine a recommended problem that matches the intent category from a pre-configured list of recommended problems; if the number of recommended problems that match the intent category is less than a first threshold, then determine the recommended problem that matches the intent category as the target problem.

[0116] In the above implementation, the question determination module 630 can also be used to determine, from the recommended questions matching the intent category, a recommended question whose intent score satisfies the target score condition as the target question if the number of recommended questions matching the intent category is greater than or equal to the first threshold. The intent score is based on the historical feedback data of the recommended question.

[0117] In some embodiments, the information recommendation device 600 may further include a question generation module. The question generation module is used to: acquire sample dialogue data from different historical dialogue scenarios; utilize the intent recognition model and, based on the sample dialogue data, determine user intent information in the historical dialogue scenarios as sample intent information; and generate a recommendation question matching the sample intent information using the large language model and according to the sample intent information.

[0118] In some embodiments, the information recommendation device 600 may further include a feedback data acquisition module and a question update module. The feedback data acquisition module can be used to acquire feedback data for the target question after the target question is recommended, and update the historical feedback data corresponding to the target question with the feedback data obtained this time; the question update module is used to update the recommended question that matches the user intent information if the historical feedback data meets the target feedback conditions.

[0119] In one possible implementation, the question update module can be specifically used to: if the historical feedback data satisfies the target feedback condition, and the number of recommended questions matching the user intent information is greater than or equal to a second threshold, then delete recommended questions that satisfy the target feedback condition from the recommended questions matching the user intent information.

[0120] In the above implementation, the question update module can also be used to: if the historical feedback data meets the target feedback conditions and the number of recommended questions matching the user intent information is less than the second threshold, then generate new recommended questions based on the large language model and the dialogue data, as recommended questions matching the user intent information.

[0121] In some implementations, the dialogue data includes multiple types of data, and the intent recognition module 620 can be specifically used to: extract input information corresponding to each type of data from the multiple types of data based on the information extraction model; input the input information corresponding to each type of data into the intent recognition model to obtain the intent recognition result output by the intent recognition model, which serves as the user intent information in the dialogue scenario.

[0122] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described device and module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0123] In the several embodiments provided in this application, the coupling between modules can be electrical, mechanical, or other forms of coupling.

[0124] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0125] In summary, the solution provided in this application acquires dialogue data from a dialogue scenario, utilizes an intent recognition model, and determines user intent information based on the dialogue data. It then selects a target question from a pre-configured pool of target questions that matches the user intent information. This target question is generated in advance using a large language model and based on different intent information output by the intent recognition model. Because the target question is generated using a large language model and the recognized intent information, it better aligns with the user's intent, resulting in more accurate recommendations. Furthermore, since the target question is pre-configured, there is no need to re-invoke the large language model during the dialogue process, thus improving the response speed for recommending relevant questions to the user.

[0126] Please refer to Figure 9 This document illustrates a structural block diagram of an electronic device according to an embodiment of this application. The electronic device 100 can be a smartphone, tablet computer, smartwatch, e-reader, or other electronic device capable of running applications. The electronic device 100 in this application may include one or more of the following components: a processor 110, a memory 120, and one or more applications, wherein the one or more applications can be stored in the memory 120 and configured to be executed by the one or more processors 110, and the one or more applications are configured to perform the methods described in the foregoing method embodiments.

[0127] Processor 110 may include one or more processing cores. Processor 110 connects to various parts within the electronic device 100 using various interfaces and lines, and performs various functions and processes data of the electronic device 100 by running or executing instructions, programs, code sets, or instruction sets stored in memory 120, and by calling data stored in memory 120. Optionally, processor 110 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). Processor 110 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into processor 110 and may be implemented separately using a communication chip.

[0128] The memory 120 may include random access memory (RAM) or read-only memory (ROM). The memory 120 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 120 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), and instructions for implementing the various method embodiments described below. The data storage area may also store data created by the electronic device 100 during use (such as phonebook data, audio and video data, chat log data, etc.).

[0129] Please refer to Figure 10 This diagram illustrates a structural block diagram of a computer-readable storage medium provided in an embodiment of this application. The computer-readable medium 800 stores program code that can be called by a processor to execute the methods described in the above method embodiments.

[0130] The computer-readable storage medium 800 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, the computer-readable storage medium 800 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 800 has storage space for program code 810 that performs any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code 810 may be compressed, for example, in a suitable form.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An information recommendation method, characterized in that, The method includes: Acquire dialogue data from the dialogue scenario; Using an intent recognition model and based on the dialogue data, determine the user intent information in the dialogue scenario; The recommendation question that matches the user intent information is determined from the pre-configured recommendation questions and used as the target question. The recommendation question is generated in advance through a large language model and based on different intent information output by the intent recognition model. Recommendations are made for the target problem.

2. The method according to claim 1, characterized in that, The user intent information includes intent categories. Determining a target question from pre-configured recommendation questions that matches the user intent information includes: At least one recommended question that matches the intent category is determined from the pre-configured recommended questions and used as the target question.

3. The method according to claim 2, characterized in that, The step of determining at least one recommended question matching the intent category from a pre-configured pool of recommended questions as the target question includes: Determine a recommended question that matches the intent category from a pre-configured pool of recommended questions; If the number of recommended questions matching the intent category is less than a first threshold, then the recommended question matching the intent category is determined as the target question.

4. The method according to claim 3, characterized in that, The step of determining at least one recommended question matching the intent category from a pre-configured pool of recommended questions as the target question further includes: If the number of recommended questions matching the intent category is greater than or equal to the first threshold, then a recommended question whose intent score meets the target score condition is determined from the recommended questions matching the intent category, and this is taken as the target question. The intent score is determined based on historical feedback data of the recommended question and popularity data of the recommended question.

5. The method according to claim 1, characterized in that, The recommendation question is generated in the following way: Obtain sample dialogue data from different historical dialogue scenarios; Using the intent recognition model and based on the sample dialogue data, user intent information in the historical dialogue scenario is determined as sample intent information; Using the large language model and based on the sample intent information, a recommendation question matching the sample intent information is generated.

6. The method according to any one of claims 1-5, characterized in that, After making recommendations for the target problem, the method further includes: Obtain feedback data for the target problem, and update the historical feedback data corresponding to the target problem with the feedback data obtained this time; If the historical feedback data meets the target feedback conditions, the recommendation question matching the user intent information is updated.

7. The method according to claim 6, characterized in that, If the historical feedback data meets the target feedback conditions, the recommendation question matching the user intent information is updated, including: If the historical feedback data meets the target feedback condition, and the number of recommended questions that match the user intent information is greater than or equal to the second threshold, then recommended questions that meet the target feedback condition are deleted from the recommended questions that match the user intent information.

8. The method according to claim 7, characterized in that, The step of updating the recommendation question that matches the user intent information if the historical feedback data meets the target feedback conditions also includes: If the historical feedback data meets the target feedback conditions, and the number of recommended questions matching the user intent information is less than the second threshold, then a new recommended question is generated based on the large language model and the dialogue data, which is used as the recommended question matching the user intent information.

9. The method according to any one of claims 1-5, characterized in that, The dialogue data includes various types of data. The process of using an intent recognition model and determining user intent information in the dialogue scenario based on the dialogue data includes: Based on the information extraction model, the input information corresponding to each type of data is extracted from the various types of data; The input information corresponding to each type of data is input into the intent recognition model to obtain the intent recognition result output by the intent recognition model, which serves as the user intent information in the dialogue scenario.

10. An information recommendation device, characterized in that, The device includes: a data acquisition module, an intent recognition module, a question determination module, and a question recommendation module, wherein, The data acquisition module is used to acquire dialogue data in the dialogue scenario; The intent recognition module is used to determine user intent information in the dialogue scenario based on the intent recognition model and the dialogue data. The problem determination module is used to determine a recommended question that matches the user intent information from a pre-configured list of recommended questions, which serves as the target question. The recommended question is generated in advance through a large language model and based on different intent information output by the intent recognition model. The problem recommendation module is used to recommend solutions to the target problem.

11. An electronic device, characterized in that, include: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform the method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains program code that can be invoked by a processor to execute the method as described in any one of claims 1-9.

Images