A multi-task intelligent question-answering method and system for a chat robot

Abstract

The application discloses a kind of multi-task intelligent question and answer method and system for chat robot, it is related to data processing technical field, the method includes: obtaining target chat robot;Obtain the multiple task types of multiple question and answer tasks proposed to target chat robot currently, obtain multiple matching sub-sample data sets, and calculate and obtain multiple matching data scale coefficients;Obtain multiple answer accuracy coefficients in past time;Calculate and obtain multiple answer recommendation coefficients;Determine whether multiple answer recommendation coefficients meet preset recommendation coefficient threshold;The answer recommendation coefficient meeting preset recommendation coefficient threshold is input into answer recommendation display analysis model, and multiple answer results are displayed using multiple answer recommendation display schemes.The application solves the technical problems of low intelligent degree and low feedback efficiency of multi-task intelligent question and answer feedback of chat robot in the prior art, and achieves the technical effect of improving the accuracy of answer results.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a multi-task intelligent question-answering method and system for chatbots. Background Technology

[0002] With the rapid development of economy and science and technology, chatbot technology has received widespread attention and is being updated and iterated rapidly to provide people with high-quality robot services.

[0003] Currently, chatbots can be further categorized into two types: goal-driven chatbots, where the target audience is defined and therefore the responses fall within a consistent domain; and goalless chatbots, where the target audience and service domain are uncertain and the range of content to be processed is broader. Both goal-driven and goalless chatbots face challenges in multi-task question-and-answer processing, including low accuracy and long analysis and feedback cycles, resulting in a poor user experience. Existing technologies suffer from low levels of intelligence and efficiency in multi-task intelligent question-and-answer feedback in chatbots. Summary of the Invention

[0004] This application provides a multi-task intelligent question-answering method and system for chatbots, which addresses the technical problems of low intelligence and low efficiency in the multi-task intelligent question-answering feedback of existing chatbots.

[0005] In view of the above problems, this application provides a multi-task intelligent question answering method and system for chatbots.

[0006] A first aspect of this application provides a multi-task intelligent question-answering method for a chatbot, the method comprising:

[0007] Obtain a target chatbot, wherein the target chatbot is trained based on a sample dataset as construction data, and the sample dataset includes multiple sub-sample datasets of multiple data types.

[0008] Obtain multiple task types of multiple question-answering tasks currently proposed to the target chatbot, match the multiple task types with the multiple subsample datasets to obtain multiple matched subsample datasets, and calculate multiple matching data scale coefficients.

[0009] Based on the answer data of the target chatbot over a past period, obtain multiple answer accuracy coefficients for the multiple task types over a past period;

[0010] Based on the multiple matching data scale coefficients and the multiple answer accuracy coefficients, multiple answer recommendation coefficients for the multiple question-answering tasks are calculated.

[0011] Determine whether the multiple answer recommendation coefficients meet the preset recommendation coefficient threshold, sort the answer recommendation coefficients that meet the thresholds in descending order, and input the corresponding question-and-answer task into the target chatbot to obtain multiple answer results;

[0012] The recommendation coefficients of answers that meet the preset recommendation coefficient threshold are input into the answer recommendation display analysis model to obtain multiple answer recommendation display schemes, and the multiple answer recommendation display schemes are used to display the multiple answer results.

[0013] A second aspect of this application provides a multi-task intelligent question-answering system for chatbots, the system comprising:

[0014] A robot acquisition module is used to acquire a target chatbot, wherein the target chatbot is acquired by training based on a sample dataset as construction data, and the sample dataset includes multiple sub-sample datasets of multiple data types.

[0015] The sample data acquisition module is used to acquire multiple task types of multiple question-and-answer tasks currently proposed to the target chatbot, match the multiple task types with the multiple sub-sample datasets to obtain multiple matched sub-sample datasets, and calculate multiple matching data scale coefficients.

[0016] An accuracy coefficient acquisition module is used to obtain multiple answer accuracy coefficients for the multiple task types over a past period based on the answer data of the target chatbot over a past period.

[0017] The recommendation coefficient calculation module is used to calculate multiple answer recommendation coefficients for the multiple question-and-answer tasks based on the multiple matching data scale coefficients and the multiple answer accuracy coefficients.

[0018] The answer result acquisition module is used to determine whether the recommendation coefficients of the multiple answers meet the preset recommendation coefficient threshold, sort the matching answer recommendation coefficients in descending order, and input the corresponding question and answer task into the target chatbot to obtain multiple answer results;

[0019] The result display module is used to input the answer recommendation coefficients that meet the preset recommendation coefficient threshold into the answer recommendation display analysis model to obtain multiple answer recommendation display schemes, and to display the multiple answer results using the multiple answer recommendation display schemes.

[0020] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0021] This application obtains a target chatbot, which is trained based on a sample dataset. This sample dataset includes multiple sub-sample datasets of various data types. Then, it acquires multiple task types for multiple question-and-answer tasks currently posed to the target chatbot, matches these task types with the sub-sample datasets to obtain multiple matched sub-sample datasets, and calculates multiple matching data size coefficients. Based on the target chatbot's past response data, it obtains multiple response accuracy coefficients for multiple task types over a past period. Then, based on the multiple matching data size coefficients and multiple response accuracy coefficients, it calculates multiple response recommendation coefficients for multiple question-and-answer tasks. It then determines whether the multiple response recommendation coefficients meet a preset recommendation coefficient threshold, sorts the matching response recommendation coefficients in descending order, and inputs the corresponding question-and-answer tasks into the target chatbot to obtain multiple response results. Finally, it inputs the response recommendation coefficients that meet the preset recommendation coefficient threshold into an response recommendation display analysis model to obtain multiple response recommendation display schemes, and displays the multiple response results using these schemes. This achieves the technical effect of improving the intelligence level of response result display and increasing the accuracy of chatbot responses. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A schematic flowchart of a multi-task intelligent question-answering method for a chatbot is provided in an embodiment of this application;

[0024] Figure 2 This application provides a flowchart illustrating a multi-task intelligent question-answering method for chatbots, which obtains multiple task types.

[0025] Figure 3A flowchart illustrating the process of obtaining multiple answer accuracy coefficients in a multi-task intelligent question-answering method for a chatbot, provided in an embodiment of this application;

[0026] Figure 4 This is a schematic diagram of a multi-task intelligent question-answering system for a chatbot, provided as an embodiment of this application.

[0027] Figure labeling: Robot acquisition module 11, Sample data acquisition module 12, Accuracy coefficient acquisition module 13, Recommendation coefficient calculation module 14, Response result acquisition module 15, Result display module 16. Detailed Implementation

[0028] This application provides a multi-task intelligent question-answering method for chatbots, which addresses the technical problems of low intelligence and low efficiency in the multi-task intelligent question-answering feedback of existing chatbots.

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0030] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product, or device.

[0031] Example 1

[0032] like Figure 1 As shown, this application provides a multi-task intelligent question-answering method for chatbots, the method comprising:

[0033] Step S100: Obtain the target chatbot, wherein the target chatbot is trained based on a sample dataset as construction data, and the sample dataset includes multiple sub-sample datasets of multiple data types;

[0034] Specifically, the target chatbot is a program designed to simulate human dialogue or conversation. It is obtained by acquiring a large number of dialogue or chat samples as a sample dataset, and then using this dataset for supervised training of the network structure. The sample dataset includes multiple sub-sample datasets of various data types. These data types simulate the user's output language, including text, speech, images, and video. The sub-sample datasets are based on different chat purposes, such as sub-sample datasets for asking about the weather, dates, and times. By training the target chatbot using these sample datasets, it can understand the user's natural language and provide corresponding responses, achieving multi-functional automatic replies. The process of training the target chatbot is existing technology.

[0035] For example, chatbot applications are closely related to daily life. For instance, they can be used for intelligent customer service by proactively sending messages and providing pre-set intelligent responses. Through basic semantic understanding and information retrieval, they can allow users to directly answer questions, improving information retrieval efficiency and shortening response cycles. For different application scenarios, multiple sub-sample datasets of various data types are obtained for each scenario. These are then aggregated to generate the sample dataset, providing training data for training the target chatbot.

[0036] Step S200: Obtain multiple task types of multiple question-answering tasks proposed to the target chatbot, match the multiple task types with the multiple subsample datasets to obtain multiple matching subsample datasets, and calculate multiple matching data scale coefficients.

[0037] Furthermore, such as Figure 2 As shown, to obtain multiple task types for multiple question-and-answer tasks currently proposed to the target chatbot, step S200 of this application embodiment further includes:

[0038] Step S210: Obtain the domain types included in the multiple question-answering tasks, and obtain multiple domain type information;

[0039] Step S220: Obtain the question and answer types included in the multiple question and answer tasks, and obtain multiple question and answer type information;

[0040] Step S230: Combine the multiple domain type information and the multiple question-and-answer type information in a one-to-one correspondence to obtain the multiple task types.

[0041] Furthermore, the multiple task types and the multiple subsample datasets are matched to obtain multiple matching subsample datasets, and multiple matching data scale coefficients are calculated. Step S200 in this embodiment further includes:

[0042] Step S240: Obtain the task types for training the target chatbot using the multiple subsample datasets, and obtain information on multiple training task types;

[0043] Step S250: Match the multiple task types and the multiple training task type information, and use the subsample datasets corresponding to the matching training task type information as matching subsample datasets to obtain multiple matching subsample datasets;

[0044] Step S260: Calculate the ratio of the amount of data in the multiple matching subsample datasets to the amount of data in the sample dataset, and obtain the multiple matching data scale coefficients.

[0045] Specifically, the process involves collecting data on multiple question-and-answer tasks posed by users to the target chatbot at any given moment—that is, collecting data on the target chatbot's tasks. These task types are then matched with multiple subsample datasets to obtain multiple matching subsample datasets. In other words, by extracting keywords from the multiple task types, the extracted keywords that accurately summarize each task type are matched with data in the multiple subsample datasets. If a match is successful, multiple subsample datasets corresponding to the multiple task types can be obtained.

[0046] Specifically, the multiple question-and-answer tasks are extracted using domain type as a keyword to obtain multiple domain type information. This domain type information reflects the domain to which the multiple question-and-answer tasks belong, including fields such as machinery, electronic communications, and customer service. Data extraction is then performed on the multiple question-and-answer tasks using question-and-answer type as a keyword to obtain multiple question-and-answer type information. This information describes the type corresponding to each question-and-answer task, including the question-and-answer purpose and method. The question-and-answer purpose refers to the desired outcome after querying the target chatbot, including recipe search results, route search results, and product search results. The question-and-answer method refers to the way the target chatbot is queried, such as through text input or voice query.

[0047] Specifically, the multiple domain type information and the multiple question-and-answer type information are mapped one-to-one. In the process of mapping one-to-one, the objects extracted by keywords are mapped one-to-one to obtain the multiple task types.

[0048] Specifically, by determining the task types within the multiple subsample datasets—that is, the chat task types included in the subsample datasets, such as querying recipes or searching routes between two locations—the purpose of training the target chatbot using these multiple subsample datasets is identified, i.e., the task types. The analyzed task types are then summarized to obtain the multiple training task type information. This multiple training task type information reflects the objectives that can be achieved after training the target chatbot using the multiple subsample datasets.

[0049] Specifically, using type keywords as the extraction target, information on the multiple task types and the multiple training task types is extracted respectively. The extraction results are matched to obtain the information on the multiple training task types that are successfully matched. Then, based on the correspondence between the information on the multiple training task types and the multiple subsample datasets, multiple matching subsample datasets corresponding to the multiple task types are determined.

[0050] Specifically, after obtaining the multiple matching subsample datasets, the number of data points in each of the multiple matching subsample datasets is counted, and this count is taken as the data volume within the multiple matching subsample datasets. The ratio of the data volume within the multiple matching subsample datasets to the data volume within the sample dataset is set as the multiple matching data size coefficient. The multiple matching data size coefficients reflect the proportion of the data volume in each matching subsample dataset to the data volume in the sample dataset. A larger coefficient indicates a higher proportion, a larger data volume in the corresponding subsample dataset, and thus higher accuracy and reliability in training the target chatbot using that subsample dataset.

[0051] Step S300: Based on the answer data of the target chatbot over a past time period, obtain multiple answer accuracy coefficients for the multiple task types over a past time period;

[0052] Furthermore, such as Figure 3 As shown, based on the answer data of the target chatbot over a past period, multiple answer accuracy coefficients for the multiple task types are obtained over a past period. In this embodiment, step S300 further includes:

[0053] Step S310: Obtain a historical question-and-answer task dataset based on the data of the target chatbot's question-and-answer tasks within a preset time range in the past;

[0054] Step S320: Extract the question-and-answer task data in the historical question-and-answer task dataset according to the multiple task types to obtain multiple historical sub-question-and-answer task datasets. Each historical sub-question-and-answer task dataset includes multiple question-and-answer tasks and answer results.

[0055] Step S330: Determine whether the answer results in the multiple historical sub-question-answering task datasets are accurate, and obtain multiple judgment result sets;

[0056] Step S340: Calculate the proportion of "yes" judgment results in the multiple judgment result sets respectively, and obtain the multiple answer accuracy coefficients.

[0057] Specifically, using the target chatbot's answer data tags as an index, answer data is searched within the chatbot's operation log. During the search, a preset time range (e.g., half a month, a quarter) is set, and data is filtered and searched from a large dataset to obtain data on question-and-answer tasks performed within the preset time period. This data is then designated as the historical question-and-answer task dataset. The historical question-and-answer task dataset comprises data generated by the target chatbot's question-and-answer responses within a historical time period, including information such as answer type, answer content, answer time, and question-and-answer task.

[0058] Specifically, by using the multiple task types as extraction labels, data is extracted from the question-and-answer task data within the historical question-and-answer task dataset, thereby obtaining multiple historical sub-question-and-answer task datasets that conform to the multiple task types. In other words, based on the determined multiple task types to be answered, data dimensionality reduction is performed from the numerous question-and-answer task data within the historical question-and-answer task dataset according to the multiple task types to find multiple historical sub-question-and-answer task datasets that meet the requirements. Each of these multiple historical sub-question-and-answer task datasets includes multiple question-and-answer tasks and answer results. Each question-and-answer task corresponds one-to-one with its answer result.

[0059] Preferably, the multiple judgment result sets are obtained by performing semantic understanding on the user's feedback evaluation of the answer results, and using the semantic understanding results as the evaluation source for whether the answer results are accurate. These multiple judgment result sets are the results obtained after evaluating the accuracy of the answer results.

[0060] Specifically, by performing a proportional analysis on each of the multiple judgment result sets, the proportion of "yes" judgments in each set is analyzed, and these proportions are used as multiple answer accuracy coefficients. A higher multiple answer accuracy coefficient indicates that the target chatbot's answers to the sub-question-answering task are more accurate. Thus, the technical effect of determining the accuracy of the target chatbot's answers to different task types based on historical data is achieved.

[0061] Step S400: Calculate multiple answer recommendation coefficients for the multiple question-and-answer tasks based on the multiple matching data scale coefficients and the multiple answer accuracy coefficients;

[0062] Specifically, based on the multiple matching data scale coefficients and the multiple answer accuracy coefficients, the recommendation degree of different answers to the multiple question-and-answer tasks is quantitatively calculated to obtain the multiple answer recommendation coefficients. Preferably, the larger the multiple answer recommendation coefficients, the higher the recommendation degree of the corresponding answer. Thus, the goal of recommending answers from two dimensions—data scale and answer accuracy—is achieved.

[0063] Optionally, multiple experts are isolated, and each expert evaluates the contribution of the matching data size coefficient and the answer accuracy coefficient to the answer recommendation coefficient. The average of these evaluation results is then calculated to obtain the weighted values ​​for the matching data size coefficient and the answer accuracy coefficient in calculating the answer recommendation coefficient. The multiple matching data size coefficients and the multiple answer accuracy coefficients are then weighted according to these weighted values ​​to obtain the multiple answer recommendation coefficients.

[0064] Step S500: Determine whether the multiple answer recommendation coefficients meet the preset recommendation coefficient threshold, sort the qualified answer recommendation coefficients in descending order, and input the corresponding question-and-answer task into the target chatbot to obtain multiple answer results;

[0065] Furthermore, determining whether the multiple answer recommendation coefficients meet a preset recommendation coefficient threshold, and sorting the matching answer recommendation coefficients in descending order, step S500 in this embodiment of the application further includes:

[0066] Step S510: Based on the answer data of the target chatbot over a past time period, set the threshold for obtaining the preset recommendation coefficient;

[0067] Step S520: Determine whether the recommendation coefficients of the multiple answers are greater than or equal to the preset recommendation coefficient threshold;

[0068] Step S530: Sort the recommendation coefficients of the answers that are judged as "yes" in descending order, and input the corresponding question-and-answer task into the target chatbot to obtain the multiple answer results.

[0069] Specifically, the preset recommendation coefficient threshold is a pre-set minimum recommendation coefficient for an answer to be recommended. This is used to filter the multiple answer recommendation coefficients, eliminating those that do not meet the minimum requirements, improving processing efficiency, and avoiding the output of answers that are likely incorrect. The preset recommendation coefficient threshold is obtained by analyzing the target chatbot's answer data over a past period to obtain the recommendation coefficients corresponding to answers that meet the user's minimum requirements. For example, based on the method described above, multiple historical answer recommendation coefficients from past answer data can be obtained, and their average can be calculated as the preset recommendation coefficient threshold.

[0070] Specifically, based on the preset recommendation coefficient threshold, the recommendation coefficients of the multiple answers are filtered. Answers with a "yes" result (i.e., those meeting the preset recommendation coefficient threshold) are sorted in descending order. The corresponding question-and-answer task is then input into the target chatbot according to the sorting result, thereby obtaining the multiple answer results. These multiple answer results represent responses to the question-and-answer task.

[0071] Step S600: Input the answer recommendation coefficients that meet the preset recommendation coefficient threshold into the answer recommendation display analysis model to obtain multiple answer recommendation display schemes, and display the multiple answer results using the multiple answer recommendation display schemes.

[0072] Furthermore, the answer recommendation coefficients that meet the preset recommendation coefficient threshold are input into the answer recommendation display analysis model to obtain multiple answer recommendation display schemes. Step S600 in this embodiment further includes:

[0073] Step S610: Based on the answer data of the target chatbot over a past period, analyze and obtain recommendation coefficients for multiple sample answers;

[0074] Step S620: Based on the recommendation coefficients of the multiple sample answers, construct multiple sample answer recommendation display schemes;

[0075] Step S630: Using the multiple sample answer recommendation coefficients and the multiple sample answer recommendation display schemes as construction data, construct the answer recommendation display analysis model;

[0076] Step S640: Input the answer recommendation coefficients that meet the preset recommendation coefficient thresholds into the answer recommendation display analysis model to obtain the multiple answer recommendation display schemes.

[0077] Specifically, the answer recommendation display analysis model is a functional model that intelligently analyzes multiple answers based on their recommendation coefficients. The multiple answer recommendation display scheme is an operational method for intelligently displaying the obtained multiple answer results. For example, the multiple answer results are displayed sequentially in descending order of recommendation coefficient, with the first third of the answer results displayed for 30 seconds, the next third for 15 seconds (sufficient time for user browsing), and the remaining answer results displayed for 20 seconds.

[0078] For example, the multiple answer results are displayed in descending order of recommendation coefficient, with the first third of the answer results occupying 50% of the display area, the middle third occupying 30% of the display area, and the last third occupying 20% ​​of the display area.

[0079] Specifically, recommendation coefficients for multiple sample answers are calculated based on the target chatbot's past response data. These recommendation coefficients reflect the degree to which each answer is recommended. Staff then construct various sample answer recommendation display schemes based on these coefficients. These schemes represent different types of display methods for the answer data.

[0080] Specifically, the recommendation coefficients of the multiple sample answers and the multiple sample answer recommendation display schemes are used as the construction data for building the answer recommendation display analysis model, and the model is trained and validated. The input data of the answer recommendation display analysis model is the answer recommendation coefficients that meet the preset advancement coefficient threshold, and the output data is the multiple answer recommendation display schemes. This achieves the goal of laying the groundwork for building an intelligent answer recommendation display analysis model.

[0081] Furthermore, using the multiple sample answer recommendation coefficients and the multiple sample answer recommendation display schemes as construction data, the answer recommendation display analysis model is constructed. In this embodiment, step S630 further includes:

[0082] Step S631: Based on the BP neural network, construct the network structure of the answer recommendation display analysis model. The input data of the answer recommendation display analysis model is the answer recommendation coefficient, and the output data is the answer recommendation display scheme.

[0083] Step S632: Perform data annotation and partitioning on the multiple sample answer recommendation coefficients and the multiple sample answer recommendation display schemes to obtain training set, validation set and test set;

[0084] Step S633: Use the training set to perform iterative supervised training on the answer recommendation display analysis model, and continuously update the network parameters of the answer recommendation display analysis model until the convergence condition is met;

[0085] Step S634: Validate and test the answer recommendation display analysis model using the validation set and test set. If the accuracy meets the preset requirements, the answer recommendation display analysis model is obtained.

[0086] Specifically, the network structure of the answer recommendation and display analysis model is constructed based on a backpropagation (BP) neural network. The constructed data is divided according to a certain ratio, and multiple sample answer recommendation display schemes are labeled. Preferably, the partition ratios of the validation set and the test set are the same, while the partition ratio of the training set is greater than that of the validation set and the test set, such as setting the partition ratio to 4:3:3. The answer recommendation and display analysis model is iteratively trained with supervision using the training set, and the training process is supervised according to the labels until the model converges.

[0087] Specifically, the recommendation coefficients of multiple sample answers from the validation set are input into the convergent answer recommendation display analysis model, and multiple validation sample answer recommendation display schemes are input. The multiple validation sample answer recommendation display schemes are matched with the multiple sample answer recommendation display schemes in the validation set, and the proportion of successful matches is set as the accuracy rate. When the accuracy rate meets the requirements, the model passes the validation.

[0088] Specifically, the data in the test set is input into the answer recommendation display analysis model, the model's computation time is statistically analyzed, and the model's running speed is judged based on the statistical results to determine whether it meets the requirements. If it does, the completed answer recommendation display analysis model is then input into the test.

[0089] In summary, the embodiments of this application have at least the following technical effects:

[0090] This application obtains the question-and-answer tasks that a chatbot needs to complete, using these tasks as the target of analysis. It then matches these tasks with multiple sub-sample datasets, calculating multiple matching data scale coefficients to quantify the size of each task within the dataset. Next, using the chatbot's past response data, it calculates the accuracy of the chatbot's responses to different task types over a past period. Based on the calculation results and the multiple matching data scale coefficients, it obtains multiple answer recommendation coefficients, providing a basis for ranking the answers. Finally, it uses an intelligent answer recommendation display analysis model to analyze these coefficients, resulting in multiple answer recommendation display schemes. This achieves the goal of intelligently analyzing the display of multi-task question-and-answer scenarios, thereby improving the feedback efficiency of multi-task question-and-answer systems.

[0091] Example 2

[0092] Based on the same inventive concept as the multi-task intelligent question-answering method for chatbots in the foregoing embodiments, such as Figure 4 As shown, this application provides a multi-task intelligent question-answering system for chatbots. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0093] Robot acquisition module 11, the robot acquisition module 11 is used to acquire target chatbot, wherein the target chatbot is acquired by training based on sample dataset as construction data, wherein the sample dataset includes multiple sub-sample datasets of multiple data types;

[0094] The sample data acquisition module 12 is used to acquire multiple task types of multiple question-and-answer tasks currently proposed to the target chatbot, match the multiple task types with the multiple sub-sample datasets to obtain multiple matching sub-sample datasets, and calculate multiple matching data scale coefficients.

[0095] Accuracy coefficient acquisition module 13 is used to obtain multiple answer accuracy coefficients of the multiple task types in the past time based on the answer data of the target chatbot in the past time.

[0096] The recommendation coefficient calculation module 14 is used to calculate multiple answer recommendation coefficients for the multiple question-and-answer tasks based on the multiple matching data scale coefficients and the multiple answer accuracy coefficients.

[0097] The answer result acquisition module 15 is used to determine whether the multiple answer recommendation coefficients meet the preset recommendation coefficient threshold, sort the qualified answer recommendation coefficients in descending order, and input the corresponding question and answer task into the target chatbot to obtain multiple answer results;

[0098] The result display module 16 is used to input the answer recommendation coefficients that meet the preset recommendation coefficient threshold into the answer recommendation display analysis model to obtain multiple answer recommendation display schemes, and to display the multiple answer results using the multiple answer recommendation display schemes.

[0099] Furthermore, the system also includes:

[0100] A type information acquisition unit is used to acquire the domain types included in the multiple question-answering tasks and acquire multiple domain type information.

[0101] A question-and-answer type information acquisition unit is used to acquire the question-and-answer types included in the multiple question-and-answer tasks and acquire multiple question-and-answer type information.

[0102] The task type acquisition unit is used to combine the multiple domain type information and the multiple question-and-answer type information in a one-to-one correspondence to obtain the multiple task types.

[0103] Furthermore, the system also includes:

[0104] A training task information acquisition unit is used to acquire the task types of the multiple subsample datasets used to train the target chatbot, and to acquire multiple training task type information.

[0105] A matching sample dataset acquisition unit is used to perform matching using the multiple task types and the multiple training task type information, and to use the sub-sample datasets corresponding to the matched training task type information as matching sub-sample datasets to obtain multiple matching sub-sample datasets.

[0106] The scale coefficient obtaining unit is used to calculate the ratio of the amount of data in the plurality of matching subsample datasets to the amount of data in the sample dataset, and obtain the scale coefficient of the plurality of matching data.

[0107] Furthermore, the system also includes:

[0108] The task dataset acquisition unit is used to obtain a historical question-and-answer task dataset based on the data of the target chatbot's question-and-answer tasks within a preset time range in the past.

[0109] Multiple historical task acquisition units are used to extract question and answer task data in the historical question and answer task dataset according to the multiple task types to obtain multiple historical sub-question and answer task datasets. Each historical sub-question and answer task dataset includes multiple question and answer tasks and answer results.

[0110] The judgment result set acquisition unit is used to judge whether the answer results in the multiple historical sub-question answering task datasets are accurate, and to obtain multiple judgment result sets;

[0111] An accuracy coefficient acquisition unit is used to calculate the proportion of "yes" judgment results in the plurality of judgment result sets, and obtain the plurality of answer accuracy coefficients.

[0112] Furthermore, the system also includes:

[0113] A recommendation coefficient threshold setting unit is used to set the preset recommendation coefficient threshold based on the answer data of the target chatbot over a past time period.

[0114] A preset recommendation coefficient threshold acquisition unit is used to determine whether the recommendation coefficients of the plurality of answers are greater than or equal to the preset recommendation coefficient threshold.

[0115] Multiple answer result acquisition units are configured to sort the recommendation coefficients of answers with a judgment result of "yes" in descending order, and input the corresponding question-and-answer task into the target chatbot to obtain the multiple answer results.

[0116] Furthermore, the system also includes:

[0117] A sample recommendation coefficient acquisition unit is used to analyze and obtain multiple sample answer recommendation coefficients based on the answer data of the target chatbot over a past period.

[0118] A recommendation display scheme construction unit is used to construct multiple sample answer recommendation display schemes based on the multiple sample answer recommendation coefficients;

[0119] The display analysis model construction unit is used to construct the answer recommendation display analysis model by using the multiple sample answer recommendation coefficients and the multiple sample answer recommendation display schemes as construction data.

[0120] The answer recommendation display scheme acquisition unit is used to input the answer recommendation coefficient that meets the preset recommendation coefficient threshold into the answer recommendation display analysis model to obtain the multiple answer recommendation display schemes.

[0121] Furthermore, the system also includes:

[0122] A network structure construction unit is used to construct the network structure of the answer recommendation display analysis model based on a BP neural network. The input data of the answer recommendation display analysis model is the answer recommendation coefficient, and the output data is the answer recommendation display scheme.

[0123] A data annotation unit is used to annotate and divide the multiple sample answer recommendation coefficients and the multiple sample answer recommendation display schemes to obtain a training set, a validation set and a test set.

[0124] A supervised training unit is used to iteratively supervise the training of the answer recommendation display analysis model using the training set, and continuously update the network parameters of the answer recommendation display analysis model until the convergence condition is met.

[0125] An analysis model acquisition unit is used to verify and test the answer recommendation display analysis model using the validation set and the test set. If the accuracy meets the preset requirements, the answer recommendation display analysis model is obtained.

[0126] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0127] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0128] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A multi-task intelligent question-answering method for a chat robot, characterized by, The method comprises: obtaining a target chat robot, wherein the target chat robot is trained based on a sample data set as construction data, wherein the sample data set comprises a plurality of sub-sample data sets of multiple data types; obtaining a plurality of task types of a plurality of question and answer tasks currently proposed to the target chat robot, matching the plurality of task types with the plurality of sub-sample data sets, obtaining a plurality of matched sub-sample data sets, and calculating a plurality of matched data size coefficients; obtaining a plurality of answer accuracy coefficients of the plurality of task types within a past time according to answer data of the target chat robot within the past time; calculating a plurality of answer recommendation coefficients of the plurality of question and answer tasks according to the plurality of matched data size coefficients and the plurality of answer accuracy coefficients; determining whether the plurality of answer recommendation coefficients meet a preset recommendation coefficient threshold, sorting the answer recommendation coefficients that meet the preset recommendation coefficient threshold in descending order, and inputting the corresponding question and answer tasks into the target chat robot to obtain a plurality of answer results; inputting the answer recommendation coefficients that meet the preset recommendation coefficient threshold into an answer recommendation display analysis model to obtain a plurality of answer recommendation display schemes, and displaying the plurality of answer results using the plurality of answer recommendation display schemes; matching the plurality of task types with the plurality of sub-sample data sets to obtain a plurality of matched sub-sample data sets, and calculating a plurality of matched data size coefficients, comprising: obtaining a plurality of training task type information by training the target chat robot using the plurality of sub-sample data sets; matching the plurality of task types with the plurality of training task type information, and obtaining a plurality of matched sub-sample data sets by matching the training task type information corresponding to the sub-sample data set; calculating the ratio of the data amount in the plurality of matched sub-sample data sets to the data amount in the sample data set to obtain the plurality of matched data size coefficients; inputting the answer recommendation coefficients that meet the preset recommendation coefficient threshold into an answer recommendation display analysis model to obtain a plurality of answer recommendation display schemes, comprising: obtaining a plurality of sample answer recommendation coefficients by analyzing the answer data of the target chat robot within a past time; constructing a plurality of sample answer recommendation display schemes based on the plurality of sample answer recommendation coefficients; constructing the answer recommendation display analysis model using the plurality of sample answer recommendation coefficients and the plurality of sample answer recommendation display schemes as construction data; inputting the answer recommendation coefficients that meet the preset recommendation coefficient threshold into the answer recommendation display analysis model to obtain the plurality of answer recommendation display schemes.

2. The multi-task intelligent question-answering method for chatbots according to claim 1, wherein, obtaining a plurality of task types of a plurality of question and answer tasks currently proposed to the target chat robot, comprising: obtaining a plurality of domain type information by obtaining the domain types included in the plurality of question and answer tasks; obtaining a plurality of question and answer type information by obtaining the question and answer types included in the plurality of question and answer tasks; obtaining the plurality of task types by one-to-one corresponding combination of the plurality of domain type information and the plurality of question and answer type information.

3. The multi-task intelligent question-answering method for chatbots according to claim 1, wherein, According to the answer data of the target chat robot in the past time, obtain a plurality of answer accuracy coefficients of the plurality of task types in the past time, comprising: According to the data of the target chat robot in the past preset time range, obtain the historical question and answer task data set; According to the plurality of task types, extract the question and answer task data in the historical question and answer task data set to obtain a plurality of historical sub-question and answer task data sets, each of which includes a plurality of question and answer tasks and answer results; Judge whether the answer results in the plurality of historical sub-question and answer task data sets are accurate to obtain a plurality of judgment result sets; Calculate the proportion of the judgment results in the plurality of judgment result sets respectively to obtain the plurality of answer accuracy coefficients.

4. The multi-task intelligent question-answering method for chatbots according to claim 1, wherein, Judge whether the plurality of answer recommendation coefficients meet the preset recommendation coefficient threshold, sort the answer recommendation coefficients that meet the preset recommendation coefficient threshold in descending order, comprising: According to the answer data of the target chat robot in the past time, set the preset recommendation coefficient threshold; Judge whether the plurality of answer recommendation coefficients are greater than or equal to the preset recommendation coefficient threshold; According to the plurality of answer recommendation coefficients that meet the preset recommendation coefficient threshold, sort the answer recommendation coefficients in descending order, and input the corresponding question and answer task into the target chat robot to obtain the plurality of answer results.

5. The multi-task intelligent question-answering method for chatbots according to claim 4, wherein, Use the plurality of sample answer recommendation coefficients and the plurality of sample answer recommendation display schemes as construction data to construct the answer recommendation display analysis model, comprising: Based on the BP neural network, construct the network structure of the answer recommendation display analysis model, the input data of the answer recommendation display analysis model is the answer recommendation coefficient, and the output data is the answer recommendation display scheme; Data labeling and division are performed on the plurality of sample answer recommendation coefficients and the plurality of sample answer recommendation display schemes to obtain a training set, a verification set and a test set; Iterative supervised training is performed on the answer recommendation display analysis model using the training set, and the network parameters of the answer recommendation display analysis model are updated constantly until the convergence condition is reached; The answer recommendation display analysis model is verified and tested using the verification set and the test set, and if the accuracy meets the preset requirement, the answer recommendation display analysis model is obtained.

6. A multi-task intelligent question-answering system for chatbots, for implementing a multi-task intelligent question-answering method for chatbots according to any one of claims 1-5, characterized in that, The system comprises: The robot obtaining module is used for obtaining a target chat robot, wherein the target chat robot is trained based on a sample data set as construction data, wherein the sample data set comprises a plurality of sub-sample data sets of a plurality of data types; The sample data obtaining module is used for obtaining a plurality of task types of a plurality of question and answer tasks currently proposed to the target chat robot, matching the plurality of task types with the plurality of sub-sample data sets to obtain a plurality of matching sub-sample data sets, and calculating a plurality of matching data size coefficients; The accuracy coefficient obtaining module is used for obtaining a plurality of answer accuracy coefficients of the plurality of task types in the past time according to the answer data of the target chat robot in the past time. The recommendation coefficient calculation module is configured to calculate a plurality of answer recommendation coefficients of the plurality of question and answer tasks according to the plurality of matching data scale coefficients and the plurality of answer accuracy coefficients; The answer result obtaining module is configured to judge whether the plurality of answer recommendation coefficients meet a preset recommendation coefficient threshold, sort the answer recommendation coefficients that meet the preset recommendation coefficient threshold in descending order, input the corresponding question and answer tasks into the target chat robot, and obtain a plurality of answer results; The result display module is configured to input the answer recommendation coefficients that meet the preset recommendation coefficient threshold into an answer recommendation display analysis model, obtain a plurality of answer recommendation display schemes, and display the plurality of answer results by using the plurality of answer recommendation display schemes.

Images