Method, apparatus, device, medium and product for conversational data question and answer

Abstract

Disclosed are methods and apparatuses, devices, media, and products for conversational data question and answer. The method includes receiving input from a user, the input including a question based on a language description and indicating a data requirement. The method also includes obtaining a confidence score of a user's intent for the data requirement relative to each category using a transformer model. Based on the confidence score, determine the category to which the user's intent belongs. In response to the confidence score being lower than a threshold, determine the category to which the user's intent belongs by identifying the input using a language model. The method also includes retrieving a target knowledge item associated with the data requirement from a knowledge base based on the question and the determined category. Based on the target knowledge item, generate an answer using a language model. Visually provide an output including an answer to a user by generating a dashboard, and the output is interactive with the user. In this way, the user's intent in the language description can be fully and efficiently understood, and feedback that meets the data requirement can be accurately provided to the user.

Description

Technical Field

[0001] Embodiments of the present disclosure generally relate to the field of computers, and in particular, relate to methods and apparatuses, devices, media, and products for conversational data questioning and answering. Background Art

[0002] As a cutting-edge information retrieval and interaction method, data question answering aims to effectively meet users' data needs by using data resources, such as answering users' questions or responding to their queries. Through data question answering technology, relevant information can be found and processed in massive data, thereby providing users with valuable feedback.

[0003] With the continuous advancement of technology, data question and answer are increasingly used in various fields. For example, at the enterprise decision-making level, data question and answer technology can provide comprehensive decision-making support information and improve the efficiency and quality of decision-making. In the field of academic research, data question and answer technology can help locate the data needed for research, promote knowledge discovery and verification, and promote scientific research innovation. In addition, in the field of daily life and public services, data question and answer technology provides users with personalized question and answer services with fast responses, improves service efficiency and user experience, and brings convenience and efficiency to people's daily lives. Summary of the invention

[0004] An embodiment of the present disclosure provides a solution for conversational data question and answer.

[0005] In a first aspect of the present disclosure, a method for conversational data question and answer is provided, the method comprising receiving an input from a user, the input comprising a question based on a language description and indicating a user's data requirement for predetermined data. The method further comprises obtaining a corresponding confidence score of a user intent corresponding to the data requirement relative to each intent category in a plurality of intent categories using a transformer model. The method further comprises determining that the user intent belongs to at least one intent category in the plurality of intent categories based on the corresponding confidence score. The method further comprises determining that the user intent belongs to at least one intent category in the plurality of intent categories by identifying the input using a language model in response to the corresponding confidence score being lower than a predetermined score threshold. The method further comprises retrieving a target knowledge item associated with the data requirement from a knowledge base based on the question and the determined at least one intent category. The method further comprises generating an answer to the question using a language model based on the retrieved target knowledge item. The method further comprises visually providing an output including the answer to the user by generating a dashboard, the output being interactive with the user.

[0006] In a second aspect of the present disclosure, a device for conversational data question and answer is provided, the device comprising an input receiving module configured to receive input from a user, the input comprising a question based on a language description and indicating a user's data requirement for predetermined data. The device also comprises a confidence acquisition module configured to obtain a corresponding confidence score of a user intent corresponding to the data requirement relative to each intent category in a plurality of intent categories using a transformer model. The device also comprises a first intent determination module configured to determine that the user intent belongs to at least one intent category in a plurality of intent categories based on the corresponding confidence score. The device also comprises a second intent determination module configured to determine that the user intent belongs to at least one intent category in a plurality of intent categories by identifying the input using a language model in response to the corresponding confidence score being lower than a predetermined score threshold. The device also comprises a knowledge retrieval module configured to retrieve a target knowledge item associated with the data requirement from a knowledge base based on the question and the determined at least one intent category. The device also comprises an answer generation module configured to generate an answer to the question using a language model based on the retrieved target knowledge item. The device further includes an output providing module configured to visually provide an output including the answer to the user by generating a dashboard, where the output is interactive with the user.

[0007] According to a third aspect of the present disclosure, an electronic device is provided. The computing device includes a processor and a memory, wherein instructions are stored in the memory, and when the instructions are executed by the processor, the processor executes a method or process according to an embodiment of the present disclosure.

[0008] According to a fourth aspect of the present disclosure, a machine-readable storage medium is provided, wherein machine-executable instructions are stored on the machine-readable storage medium, and when the machine-executable instructions are executed by a processor, the processor is enabled to execute a method or process according to an embodiment of the present disclosure.

[0009] In a fifth aspect of the present disclosure, a computer program product is provided. The computer program product is tangibly stored on a non-transitory computer-readable storage medium and includes a computer program, which, when executed by a processor of a computer, causes the processor to perform a method or process according to an embodiment of the present disclosure.

[0010] Please note that the summary is provided to introduce a series of concepts in a simplified form, which will be further described in the detailed description below. The summary is not intended to identify key features or essential features of the disclosure, nor is it intended to limit the scope of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The above and other purposes, features and advantages of the present disclosure will become more clearly understood by describing the embodiments of the present disclosure in more detail in conjunction with the accompanying drawings, in which:

[0012] Figure 1 is a diagram schematically illustrating an example environment in which methods and / or processes according to embodiments of the present disclosure may be implemented;

[0013] Figure 2 is a flowchart schematically illustrating a method for conversational data question and answer according to an embodiment of the present disclosure;

[0014] Figure 3 is a diagram schematically illustrating an example process of intent type determination according to an embodiment of the present disclosure;

[0015] Figure 4 is a diagram schematically illustrating an example process of knowledge base construction and knowledge retrieval according to an embodiment of the present disclosure;

[0016] Figure 5 is a diagram schematically illustrating an example process of burying point configuration according to an embodiment of the present disclosure;

[0017] Figure 6 is a diagram schematically illustrating an example process of a conversational knowledge question and answer according to an embodiment of the present disclosure;

[0018] Figure 7 is a diagram schematically illustrating an apparatus for conversational data question and answer according to an embodiment of the present disclosure;

[0019] Figure 8 is a schematic block diagram of an example device that may be used to implement embodiments according to the present disclosure.

[0020] Throughout the drawings, same or similar reference numbers generally refer to same or similar elements. DETAILED DESCRIPTION

[0021] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0022] In the description of the embodiments of the present disclosure, the term "including" and its variations should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects, unless explicitly indicated to be different. As described above, data questioning effectively utilizes data resources to meet users' data needs. However, although data questioning is helpful, it still faces many challenges. Related data questioning solutions are usually highly dependent on users' ability to accurately and clearly describe their data needs. When the user's description is difficult and obscure, or when faced with complex and changeable data needs, the limitations of the question-and-answer system are undoubtedly highlighted.

[0023] Related data question-and-answer solutions have obvious deficiencies in deeply understanding user query intent. Their processing mechanisms lack both depth and flexibility, making it difficult to fully capture and accurately analyze user data needs. This limitation often results in users being unable to obtain satisfactory answers, and instead requiring them to invest a lot of time and energy to repeatedly adjust and optimize inputs including questions or queries in an attempt to match the system's data retrieval and processing capabilities.

[0024] In order to solve at least some of the above and other potential problems, an embodiment of the present disclosure proposes a solution for conversational data question and answer. The solution includes receiving input from a user, the input including a question based on a language description and indicating a user's data requirement for predetermined data. The solution also includes using a transformer model to obtain a corresponding confidence score of a user intent corresponding to the data requirement relative to each intent category in a plurality of intent categories. The solution also includes determining that the user intent belongs to at least one intent category in a plurality of intent categories based on the corresponding confidence score. The solution also includes determining that the user intent belongs to at least one intent category in a plurality of intent categories by identifying the input using a language model in response to the corresponding confidence score being lower than a predetermined score threshold. The solution also includes retrieving a target knowledge item associated with the data requirement from a knowledge base based on the question and at least one determined intent category. The solution also includes generating an answer to the question using a language model based on the retrieved target knowledge item. The solution also includes visually providing an output including the answer to the user by generating a dashboard, the output being interactive with the user.

[0025] In this way, through multi-model collaborative judgment, user intent can be determined more accurately and efficiently, avoiding the situation where the wrong answer is given due to misunderstanding the user's question. According to the accurate user intent, the relevant knowledge can be retrieved from the knowledge base more accurately and quickly, reducing unnecessary knowledge retrieval time and resource waste. The generated answers are visually provided to users in the form of a dashboard, and users can interact with the data results, which further enhances the personalization and interactivity of users in the process of obtaining information.

[0026] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0027] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.

[0028] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0029] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0030] Reference below Figures 1 to 8 It should be understood that these exemplary embodiments are provided only to enable those skilled in the art to better understand and implement the embodiments of the present disclosure, and are not intended to limit the scope of the present disclosure in any way.

[0031] Figure 1 1 is a schematic diagram schematically illustrating an example environment 100 in which methods and / or processes according to embodiments of the present disclosure may be implemented. Figure 1As shown in the example environment 100, it includes a user 110, a computing device 120, and a display device 130. It should be understood that limited components are shown in the example environment 100 only for the purpose of ease of understanding and ease of illustration, and the embodiments of the present disclosure are not limited thereto. For example, the example environment 100 may also include a storage device, an input device, a pointing device, etc.

[0032] The user 110 may be an individual or an organization (e.g., an enterprise). The user 110 may have a data requirement for predetermined data to be processed or analyzed. According to an embodiment of the present disclosure, the user 110 may input a question indicating his / her data requirement for predetermined data in the form of a language description, such as via a text input device or a voice input device. It should be understood that although Figure 1 Only one user is shown as an example, but the present disclosure does not limit the number of users 110, and concurrent knowledge question and answering in which multiple users ask questions at the same time can be implemented.

[0033] Computing device 120 may process input from user 110 , such as understanding the user's intent of the input and utilizing data resources (eg, stored on a storage device) to respond to the user's intent. Figure 1 The computing device 120 shown in the figure may indicate a single computing device or a collection of computing resources, which may be centrally arranged locally or distributed in the cloud. The cloud may indicate a computing and service model built on cloud or distributed technology, in which computing resources are coupled together through a network to form an intelligently scheduled and flexibly scalable collection. At least a portion of these resources may be dynamically accessed or allocated as needed to complete the corresponding calculations without the need to locally own or manage these physical resources. The various models involved in the method for conversational data question-and-answer according to an embodiment of the present disclosure may be run on the computing device 120.

[0034] Examples of computing device(s) 120 may include supercomputers, personal computers, laptop computers, vehicle-mounted computing devices, mobile devices (such as smartphones, tablet computers, etc.), wearable electronic devices, multimedia devices, personal digital assistants (PDAs), smart home devices, consumer electronics, or a combination including any one or more of the above devices, etc. It should be understood that the computing devices described herein are exemplary only and not limiting.

[0035] The display device 130 may be configured to display the generated dashboard to visually provide the output including the answer to the user. The dashboard incorporates rich data visualization elements, and the user can obtain the required information at a glance, such as through data charts, text descriptions, image displays, etc., which improves the efficiency of the user in obtaining information. In addition, the user can further interact with the output on the display device 130, such as giving further instructions, selecting specific data points, etc.

[0036] It should be understood that for ease of explanation and illustration purposes, Figure 1 Only one computing device and one display device are shown in FIG. However, this is only exemplary, and there may be other question-answering environments and systems with different configurations, which may be customized according to specific usage scenarios and needs, as is also the case below.

[0037] Combined with the above Figure 1 Describes an example environment in which the methods and / or processes according to embodiments of the present disclosure may be implemented. Figure 2 The method 200 for conversational data question and answer according to an embodiment of the present disclosure is described. Through the method 200, the user's intention in the language description can be fully and efficiently understood, and then feedback that meets the data needs is accurately provided to the user.

[0038] Figure 2 2 is a flowchart schematically illustrating a method 200 for conversational data question and answer according to an embodiment of the present disclosure. At 210, an input from a user is received, the input including a question based on a language description and indicating the user's data requirements for predetermined data. The user can ask questions in a language description manner without having to learn the syntax of the query language, the structure of the library, etc. in advance. This natural language interaction method enables non-technical users to easily obtain the required information from complex data sets, thereby improving the availability and popularity of the data question and answer system.

[0039] At 220, the transformer model is used to obtain the corresponding confidence score of the user intent corresponding to the data requirement relative to each intent category in the multiple intent categories. According to an embodiment of the present disclosure, the transformer model-based intent recognition is first performed to determine the corresponding confidence score of the user intent relative to each intent category in the multiple intent categories. The transformer model is capable of processing text of real-time conversational data questions and answers, which not only has an advantage in processing speed, but is also suitable for complex semantic understanding tasks.

[0040] According to the embodiments of the present disclosure, a large language model can be used to generate user query examples. Such user query examples can be diverse, covering data product knowledge, embedding design, dashboard generation, etc., to expand the training data set and improve the robustness and accuracy of the transformer model in practical applications. The transformer model is trained using real data and synthetic data, and the model is fine-tuned through supervised learning to adapt to specific classification tasks.

[0041] At 230, based on the corresponding confidence scores, it is determined that the user intent belongs to at least one intent category among the multiple intent categories. According to an embodiment of the present disclosure, one or more confidence scores with the highest or higher scores can be identified based on the scores of the obtained confidence scores, and based on this, the intent category corresponding to the identified one or more confidence scores is determined as the intent category to which the user intent belongs. In addition, the intent category to which the user intent belongs can also be determined based on the score interval of the obtained confidence scores. The intent category corresponding to the confidence score that falls within the predetermined score interval is determined as the intent category to which the user intent belongs.

[0042] At 240, in response to the corresponding confidence score being lower than a predetermined score threshold, the user intent is determined to belong to at least one of the multiple intent categories by using a language model to recognize the input. According to an embodiment of the present disclosure, when the obtained confidence scores are not good, intent recognition based on a language model will be performed. The language model can be a large language model (LLM), and based on an agent corresponding to the determined intent category, a question of a predetermined format is submitted to the language model to guide the model to more accurately understand and classify the user intent.

[0043] At 250, based on the question and at least one determined intent category, target knowledge items associated with the data requirement are retrieved from the knowledge base. A knowledge item may be a basic unit for storage and management in a knowledge base, which may be a sliced ​​text segment, a data table, etc. After determining the intent category of the user's intent, based on the question and the determined intent category, the associated knowledge can be retrieved from the knowledge base in a targeted and accurate manner, without traversing layer by layer to waste unnecessary knowledge retrieval time or retrieving unnecessary redundant information. In the following, the process of knowledge base construction and knowledge retrieval according to an embodiment of the present disclosure will be described in further detail.

[0044] At 260, based on the retrieved target knowledge items, the language model is used to generate answers to the questions. After the target knowledge items associated with the user's data needs are retrieved from the knowledge base, the corresponding agent is used, and the language model generates detailed and accurate answers based on the acquired knowledge according to the user's questions.

[0045] At 270, an output including the answer is visually provided to the user by generating a dashboard, and the output is interactive with the user. The dashboard integrates a variety of data visualization tools, allowing users to quickly and clearly capture the required information, thereby improving the efficiency of information acquisition. In addition, the user can also interact with these data results, which not only realizes the personalized presentation of information, but also enhances the user's interactive experience. The user can further iterate his requirements by interacting with the elements on the dashboard.

[0046] In this way, through multi-model collaborative judgment, user intent can be determined more accurately and efficiently, avoiding the situation where the wrong answer is given due to misunderstanding the user's question. According to the accurate user intent, the relevant knowledge can be retrieved from the knowledge base more accurately and quickly, reducing unnecessary knowledge retrieval time and resource waste. The generated answers are visually provided to users in the form of a dashboard, and users can interact with the data results, which further enhances the personalization and interactivity of users in the process of obtaining information.

[0047] Figure 3 3 is a diagram schematically illustrating an example process 300 of intent type determination according to an embodiment of the present disclosure. Figure 3 As shown in , input 302 indicating a user's data need can first determine the intent type of the user's intent via a transformer model 304. The transformer model 304 can calculate a corresponding confidence score of the user's intent relative to each of a plurality of intent categories, and then determine the intent category to which the user's intent belongs based on the confidence score. The plurality of intent categories can be predetermined or dynamically adjusted, for example, based on user input or usage scenarios.

[0048] If the recognition result at the transformer model 304 is sufficient to make an accurate judgment, for example, at least one of the obtained confidence scores is greater than a predetermined score threshold (i.e., "yes" at 308), the classification result 310 is determined. If the recognition result at the transformer model 304 is insufficient to make an accurate judgment, for example, the obtained confidence scores are all less than the predetermined score threshold (i.e., "no" at 308), the intent type of the user's intent can be identified via the language model 306. The collaborative intent recognition based on multiple models of the transformer model 304 and the language model 306 not only ensures the recognition efficiency, but also enables secondary confirmation when the initial classification effect is not good, and uses a large model to more accurately identify the intent category of the user's intent.

[0049] Figure 4 is a diagram schematically illustrating an example process 400 of knowledge base construction and knowledge retrieval according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the knowledge base 412 may include a plurality of vectorized knowledge items. Figure 4 Describe how these vectorized knowledge items are generated to form a data resource that can be used to respond to user input (questions, queries, etc.).

[0050] According to an embodiment of the present disclosure, a plurality of knowledge slices 408 can be obtained by slicing the knowledge document 404 using a language model 406. Different from the traditional segmentation by line breaks, such semantic segmentation can more accurately identify the document structure and semantic boundaries. After obtaining a plurality of knowledge slices 408, a corresponding question-answer pair 410 for each of the knowledge slices 408 can be determined using the language model 406 to provide direct knowledge support for subsequent queries.

[0051] Based on, for example, the number of words in the text, the complexity of the content, etc., by vectorizing the corresponding question-answer pair 410 of each knowledge slice in different dimensions, multiple vectorized knowledge items can be generated. Vectorization in different dimensions enables the requirements of different lengths and semantics to be met. These vectorized knowledge items are stored and maintained at a certain frequency to establish a basis for fast retrieval.

[0052] In order to improve the accuracy and efficiency of retrieving target knowledge items from the knowledge base 412, according to an embodiment of the present disclosure, a first number of knowledge items with a first degree of relevance are retrieved from the knowledge base 412 for user questions, such as the first ten knowledge items that are more relevant to the user question. Next, a second number of knowledge items with a second degree of relevance are selected from the first number of knowledge items using the rearrangement model 416, where the second number is less than the first number and the second degree of relevance is higher than the first degree of relevance. For example, the first five knowledge items that are more relevant to the user question are selected from the 10 knowledge items previously retrieved. After such a layer screening, the quality of the knowledge items used to generate answers can be improved, thereby improving the accuracy of the answers. Using the corresponding agent, the language model 406 can generate an output 418 including a detailed and accurate answer based on the user question and the selected second number.

[0053] According to an embodiment of the present disclosure, the rearrangement model 416 may be at least partially based on a transformer, and the training of such a rearrangement model may include retrieving a second number of training knowledge items (e.g., 5 matching fragments) from the knowledge base 412 according to synthetic data (e.g., synthetic questions 414), and generating training data for training the rearrangement model 416 by ranking the second number of training knowledge items (e.g., the degree of relevance to the synthetic questions) using the language model 406. The combination of such synthetic questions and ranked training knowledge items can ensure the performance of the rearrangement model 416 in layer screening knowledge items.

[0054] Figure 5is a diagram schematically illustrating an example process 500 of a tracking configuration according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the construction of the knowledge base 512 is based on a tracking scheme indicating a user context 504, which includes, for example, information about the user's organizational structure, and is intended to optimize data collection efficiency and effectiveness according to the user's industry background, etc.

[0055] According to an embodiment of the present disclosure, a language model 506 can be used to generate a first tracking solution 510 based on the user context 504, and the first tracking solution 510 includes multiple first trackings associated with the user. Using a corresponding agent, a language model 506 can be used to generate a standard tracking design solution based on the user's industry background, and the solution can cover common data analysis requirements, such as key indicators such as click-through rate and visit depth.

[0056] According to an embodiment of the present disclosure, a second burying point solution 514 can be generated based on the existing burying point 508 using a language model 506, and the second burying point solution 514 includes at least one second burying point that promotes the user's data needs. Using a corresponding intelligent agent, the language model 506 can be used to summarize the user's analysis trend based on the data, infer missing or potentially useful burying points, and generate a supplementary burying point design solution. Based on the standard and supplementary burying point design solutions, the constructed knowledge base 512 will be more customized and personalized for the user.

[0057] As described above, in response to receiving input 502 from the user, a first number of knowledge items related to the user's question will first be retrieved from the knowledge base 512. In this process, the design solutions that may have been adopted can also be optimized. According to an embodiment of the present disclosure, candidate solutions for burying points are determined based on retrieving a first number of knowledge items with a first degree of relevance from the knowledge base for the question. According to the user's question, a certain number of knowledge resources that match the question or query are retrieved from the knowledge base 512, and the most relevant design solution can be found based on this. In addition, a corresponding intelligent agent is used to generate an updated solution for burying points based on the candidate solution using the language model 506, and the updated solution includes burying point suggestions. The large language model can further understand the specific needs of the user, and can generate targeted and highly adaptable burying point design suggestions for the specific needs proposed by the user.

[0058] Figure 6It is a diagram schematically illustrating an example process 600 of a conversational knowledge question and answer according to an embodiment of the present disclosure. In order to generate a dashboard to visually provide an output 618 including an answer to the user, the following data processing may be performed. Event information 604 may be input into a language model 606 separately, including user-related events (such as page visits, button clicks, etc.). In addition, attribute information 608 may be input into a language model 606, including event attributes (such as time, etc.). The use of indicators 610 facilitates accurate retrieval of knowledge resources from a knowledge base 612. For example, a user-defined indicator is usually composed of an event and multiple filtering conditions, such as the buried point data of "time = xxxx year xx month xx day".

[0059] like Figure 6 As shown in , input 602 can be original input, input rewritten by approximate semantics, or input translated across languages, etc. According to an embodiment of the present disclosure, the input can be rewritten to optimize and standardize the original input of the user so as to more accurately match the data and generate relevant views. In some embodiments, by performing semantic analysis and structural reorganization on the user question using language model 606, the question can be rewritten by approximate semantics to clarify the question. In some embodiments, by using language model 606 to convert the user question from the original first language into a second language different from the first language, so as to obtain an answer in the second language to the question.

[0060] In some embodiments, the user input can be semantically expanded to ensure comprehensive understanding, such as by adding descriptive language to clarify the question or query intent. In addition, the user input can be keyword extracted to extract core vocabulary and concepts from the user's question or query, which helps the system accurately identify the data points that need to be focused on.

[0061] According to an embodiment of the present disclosure, the elements in the question are identified and filled by performing slot filling on the user question using the language model 606 for mapping to the corresponding data dimensions and parameters. By using the language model 606 in combination with the agent technology of function calls, key elements in the query, such as time range, indicators, attribute grouping, general filtering conditions and control groups, can be automatically identified and filled.

[0062] By way of example and not limitation, a user's natural language query is first received and a preliminary text parsing is performed. In combination with an agent or prompt word technique, such as using a large language model, the parsed information is mapped to predefined slots. This process involves identifying and labeling key information elements in the question or query.

[0063] In some embodiments, time alignment ensures that all time data is accurately parsed and standardized for data processing and visualization. This process involves converting time expressions in natural language into a unified format and parsing relative time expressions into specific dates. For example, regular expressions can be used to identify and convert Chinese date descriptions, such as converting "March 12, 2023" to the standard format "2023-03-12". Descriptive time words in Chinese, such as "last month", "next Monday", etc., can be parsed. This step uses vocabulary technology to determine the specific time of these words relative to the current date. All parsed dates can also be converted to timestamp format to ensure that time data can be processed uniformly in subsequent data processing and analysis.

[0064] According to an embodiment of the present disclosure, the target knowledge can be retrieved based on the demand indicator in the user question. In response to the text indicator of the retrieved knowledge item being the same as the demand indicator (i.e., "yes" at 616), the first metadata corresponding to the text indicator can be determined. In addition, in response to the text indicator of the retrieved knowledge item being different from the demand indicator (i.e., "no" at 616), at least one indicator corresponding to the user's intention is determined using the language model 606, and an additional retrieval condition, i.e., a supplementary screening condition 614, is generated.

[0065] By way of example and not limitation, the constructed knowledge base 612 can be used to recall the indicator text in the user's question or query. This step first finds the stored indicators that are most relevant to the query. If the recalled indicator text is exactly the same as the text in the user's question or query, the disambiguation process ends and the corresponding metadata is used directly. If there are differences in the text, the results of the recall are submitted to the language model 606, which will evaluate which recalled indicators are most consistent with the user's intentions and propose possible additional screening conditions 614.

[0066] According to an embodiment of the present disclosure, retrieving target knowledge can be based on the requirement attribute in the user question. In response to the text attribute of the retrieved knowledge item being the same as the requirement attribute, a second metadata corresponding to the text attribute can be determined. In response to the text attribute of the retrieved knowledge item being different from the requirement attribute, at least one attribute corresponding to the user's intention can be determined using the language model 606, and the at least one attribute includes a corresponding operator and a value to form a triple.

[0067] By way of example and not limitation, the knowledge base 612 can be used to recall the attributes mentioned in the user's question or query, and to recall the associated attribute value candidate set. For fully matched attribute texts, the corresponding metadata is used directly. For incomplete matches, the language model 606 is accessed to determine the most appropriate attribute and value combination. Each filter condition can be simplified to a triple (attribute, operator, value) to ensure accurate execution of the query.

[0068] In some embodiments, the processed results can be combined and converted into a structured query language (DSL) so that the necessary data can be retrieved from the database using a query engine and finally displayed to the user. This process ensures a seamless connection from user query to data visualization. The aforementioned processing results such as time alignment, indicator disambiguation, filter condition disambiguation, etc. are collected to form a complete query. In addition, a complete structured query language (DSL) can be constructed based on the integrated results, including time range, indicators, filter conditions, attribute grouping, etc.

[0069] Figure 7 Schematically illustrates a device 700 for conversational data question answering according to an embodiment of the present disclosure. The device 700 may include multiple units or modules for performing the steps or actions in the method or process discussed above. Figure 6 As shown in , the device 700 includes an input receiving module configured to receive input from a user, the input including a question based on a language description and indicating a user's data requirement for predetermined data. The device 700 also includes a confidence acquisition module configured to use a transformer model to obtain a corresponding confidence score of a user intent corresponding to the data requirement relative to each intent category in a plurality of intent categories. The device 700 also includes a first intent determination module configured to determine that the user intent belongs to at least one intent category in the plurality of intent categories based on the corresponding confidence score. The device 700 also includes a second intent determination module configured to determine that the user intent belongs to at least one intent category in the plurality of intent categories by identifying the input using a language model in response to the corresponding confidence score being lower than a predetermined score threshold. The device 700 also includes a knowledge retrieval module configured to retrieve a target knowledge item associated with the data requirement from a knowledge base based on the question and the determined at least one intent category. The device 700 also includes an answer generation module configured to generate an answer to the question using the language model based on the retrieved target knowledge item. The apparatus 700 further includes an output providing module configured to visually provide an output including the answer to the user by generating a dashboard, wherein the output is interactive with the user.

[0070] In some embodiments, wherein the knowledge base may include a plurality of vectorized knowledge items, the knowledge base may be constructed by: slicing the knowledge document using the language model to obtain a plurality of knowledge slices; determining a corresponding question-answer pair for each of the plurality of knowledge slices using the language model; and generating the plurality of vectorized knowledge items by vectorizing the corresponding question-answer pairs of each knowledge slice in different dimensions based on the number of words in the text and the complexity of the content.

[0071] In some embodiments, wherein the construction of the knowledge base may be based on a tracking scheme indicating a user context, wherein the user context includes information about the organizational structure of the user, the device 700 also includes a tracking scheme generation module configured to: generate a first tracking scheme based on the user context using the language model, wherein the first tracking scheme includes a plurality of first tracking points associated with the user; and generate a second tracking scheme based on existing tracking points using the language model, wherein the second tracking scheme includes at least one second tracking point that facilitates the data needs of the user.

[0072] In some embodiments, the knowledge retrieval module can be further configured to: retrieve a first number of knowledge items with a first degree of relevance from the knowledge base for the problem; and select a second number of knowledge items with a second degree of relevance from the first number of knowledge items using a rearrangement model, wherein the second number is smaller than the first number and the second degree of relevance is higher than the first degree of relevance.

[0073] In some embodiments, the rearrangement model may be at least partially based on a transformer, and the apparatus 700 further includes a training module for the rearrangement model, configured to: retrieve the second number of training knowledge items from the knowledge base according to the synthesized question; and generate training data for training the rearrangement model by sorting the second number of training knowledge items using the language model.

[0074] In some embodiments, the device 700 also includes a burying point solution update module, which is configured to: determine candidate solutions for the burying point based on retrieving the first number of knowledge items with the first degree of relevance from the knowledge base for the problem; and based on the candidate solutions, generate an updated solution for the burying point using the language model, the updated solution including burying point suggestions.

[0075] In some embodiments, the device 700 also includes a question optimization module, which is configured to: perform semantic analysis and structural reorganization on the question by utilizing the language model, and perform approximate semantic rewriting on the question to clarify the question; and convert the question from an original first language into a second language different from the first language by utilizing the language model, so as to obtain the answer to the question in the second language.

[0076] In some embodiments, the apparatus 700 further comprises a slot filling module configured to identify and fill elements in the question by performing slot filling on the question using the language model for mapping to corresponding parameters.

[0077] In some embodiments, the device 700 also includes an indicator disambiguation module, which is configured to: retrieve the target knowledge based on the requirement indicator in the question, and the method also includes: in response to the text indicator of the retrieved knowledge item being the same as the requirement indicator, determining the first metadata corresponding to the text indicator; and in response to the text indicator of the retrieved knowledge item being different from the requirement indicator, using the language model to determine at least one indicator corresponding to the user intention, and generating additional retrieval conditions.

[0078] In some embodiments, the device 700 also includes an attribute disambiguation module, which is configured to: retrieve the target knowledge based on the requirement attribute in the question, and the method also includes: in response to the text attribute of the retrieved knowledge item being the same as the requirement attribute, determining second metadata corresponding to the text attribute; and in response to the text attribute of the retrieved knowledge item being different from the requirement attribute, using the language model to determine at least one attribute corresponding to the user intention, the at least one attribute including a corresponding operator and a value to form a triple.

[0079] Figure 8 8 is a block diagram of an electronic device 800 according to some embodiments of the present disclosure. The device 800 may be a device or apparatus described in an embodiment of the present disclosure. Figure 8 As shown, the device 800 includes a central processing unit (CPU) and / or a graphics processing unit (GPU) 801, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 802 or loaded from a storage unit 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The CPU / GPU 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804. Although not shown in FIG. Figure 8As shown in FIG. 8 , device 800 may also include a co-processor.

[0080] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0081] The various methods or processes described above may be performed by the CPU / GPU 801. For example, in some embodiments, the methods may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the CPU / GPU 801, one or more steps or actions in the methods or processes described above may be performed.

[0082] In some embodiments, the methods and processes described above may be implemented as a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present disclosure.

[0083] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer readable storage medium used here is not interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.

[0084] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0085] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, and conventional procedural programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as a separate software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0086] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0087] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0088] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the equipment, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each frame in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the frame can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous frames can actually be executed substantially in parallel, and they can also be executed in the opposite order sometimes, depending on the functions involved. It should also be noted that each frame in the block diagram and / or flow chart, and the combination of frames in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0089] Various embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the various embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the various embodiments disclosed herein.

Claims

1. A method for conversational data question answering, comprising: receiving input from a user, the input comprising a question based on a language description and indicating a data requirement of the user for predetermined data; Obtaining, using a transformer model, a corresponding confidence score of a user intent corresponding to the data requirement relative to each intent category in a plurality of intent categories; Determining, based on the corresponding confidence scores, that the user intent belongs to at least one intent category among the plurality of intent categories; In response to the corresponding confidence score being lower than a predetermined score threshold, determining that the user intent belongs to the at least one intent category among the plurality of intent categories by recognizing the input using a language model; Based on the question and the determined at least one intent category, retrieving a target knowledge item associated with the data requirement from a knowledge base, comprising: For the problem, retrieving a first number of knowledge items with a first degree of relevance from the knowledge base, the first number of knowledge items being associated with candidate solutions for the embedding point, and selecting, using a rearrangement model, a second number of knowledge items having a second degree of relevance from the first number of knowledge items, the second number being smaller than the first number and the second degree of relevance being higher than the first degree of relevance; Based on the retrieved target knowledge item, generating an answer to the question using the language model; and An output including the answer is visually provided to the user by generating a dashboard, and the output is interactive with the user.

2. The method according to claim 1, wherein the knowledge base comprises a plurality of vectorized knowledge items, and the knowledge base is constructed by: By slicing the knowledge document using the language model, a plurality of knowledge slices are obtained; Determining a corresponding question-answer pair for each of the plurality of knowledge slices using the language model; and The vectorized multiple knowledge items are generated by vectorizing the corresponding question-answer pairs of each knowledge slice in different dimensions based on the number of text words and content complexity.

3. The method according to claim 2, wherein the knowledge base is constructed based on a tracking scheme indicating a user context, wherein the user context includes information about the organizational structure of the user, and the method further comprises: Generate a first tracking solution based on the user context using the language model, wherein the first tracking solution includes a plurality of first tracking points associated with the user; as well as The language model is used to generate a second embedding solution based on the existing embedding solutions, wherein the second embedding solution includes at least one second embedding solution that facilitates the data demand of the user.

4. The method of claim 1, wherein the rearrangement model is at least partially based on a transformer, and the training of the rearrangement model comprises: Retrieving the second number of training knowledge items from the knowledge base according to the synthesized question; as well as Training data for training the rearrangement model is generated by sorting the second number of training knowledge items using the language model.

5. The method according to claim 1, further comprising: Determine the candidate solution for the embedding point based on retrieving the first number of knowledge items with the first degree of relevance from the knowledge base for the problem; as well as Based on the candidate solutions, an update solution for the embedding point is generated using the language model, and the update solution includes embedding point suggestions.

6. The method according to claim 1, further comprising: By using the language model to perform semantic analysis and structural reorganization on the question, the question is approximately semantically rewritten to clarify the question; as well as The question is converted from an original first language into a second language different from the first language by using the language model, so as to obtain the answer in the second language to the question.

7. The method according to claim 1, further comprising: Elements in the question are identified and filled by performing slot filling on the question using the language model for mapping to corresponding parameters.

8. The method according to claim 1, wherein retrieving the target knowledge is based on a requirement indicator in the problem, and the method further comprises: In response to the retrieved text indicator of the knowledge item being the same as the requirement indicator, determining first metadata corresponding to the text indicator; as well as In response to the text index of the retrieved knowledge item being different from the requirement index, at least one index corresponding to the user intention is determined using the language model, and an additional retrieval condition is generated.

9. The method according to claim 1, wherein retrieving the target knowledge is based on a requirement attribute in the problem, and the method further comprises: In response to the retrieved text attribute of the knowledge item being the same as the requirement attribute, determining second metadata corresponding to the text attribute; as well as In response to the text attribute of the retrieved knowledge item being different from the requirement attribute, at least one attribute corresponding to the user intention is determined using the language model, the at least one attribute including a corresponding operator and a value to form a triple.

10. A device for conversational data question and answer, comprising: an input receiving module configured to receive input from a user, the input comprising a question based on a language description and indicating a data requirement of the user for predetermined data; A confidence acquisition module, configured to acquire, using a transformer model, a corresponding confidence score of the user intent corresponding to the data requirement relative to each intent category in a plurality of intent categories; A first intention determination module, configured to determine, based on the corresponding confidence score, that the user intention belongs to at least one of the plurality of intention categories; a second intent determination module configured to determine, in response to the corresponding confidence score being lower than a predetermined score threshold, that the user intent belongs to the at least one intent category among the plurality of intent categories by recognizing the input using a language model; A knowledge retrieval module is configured to retrieve a target knowledge item associated with the data requirement from a knowledge base based on the question and the determined at least one intent category, comprising: For the problem, retrieving a first number of knowledge items with a first degree of relevance from the knowledge base, the first number of knowledge items being associated with candidate solutions for the embedding point, and selecting, using a rearrangement model, a second number of knowledge items having a second degree of relevance from the first number of knowledge items, the second number being smaller than the first number and the second degree of relevance being higher than the first degree of relevance; an answer generation module configured to generate an answer to the question using the language model based on the retrieved target knowledge item; and The output providing module is configured to visually provide the output including the answer to the user by generating a dashboard, wherein the output is interactive with the user.

11. An electronic device comprising: processor; as well as A memory coupled to the processor, wherein instructions are stored in the memory, and when the instructions are executed by the processor, the processor is caused to perform the method according to any one of claims 1 to 9.

12. A machine-readable storage medium having machine-executable instructions stored thereon, wherein the machine-executable instructions, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 9.

13. A computer program product comprising a computer program which, when executed by a processor of a computer, causes the processor to perform the method according to any one of claims 1 to 9.

Images