Artificial intelligence-based question processing method, device, equipment and storage medium

Abstract

The present disclosure provides a question processing method and device based on artificial intelligence, equipment and storage medium, relates to the field of artificial intelligence, specifically relates to natural language processing, deep learning, intelligent question answering technology, and can be applied in smart city and smart government scenarios. The scheme is: determining a first correlation degree between a to-be-processed question and a plurality of standard answers in an answer library; determining a target answer from the plurality of standard answers according to the first correlation degrees of the plurality of standard answers; and replying to the to-be-processed question according to the target answer. Thus, the correlation degree between the to-be-processed question and the plurality of standard answers in the answer library can be used to determine a reply answer (i.e., the target answer) corresponding to the to-be-processed question from the plurality of standard answers, and the accuracy of determining the target answer can be improved. In addition, in the scene of manual reply by customer service personnel, the determined target answer can be provided to the customer service personnel to assist the customer service personnel in quickly solving user problems and improving the service efficiency of the customer service personnel.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of artificial intelligence, in particular to natural language processing, deep learning, intelligent question answering technology, which can be applied in smart city and smart government scenarios, and more particularly to an artificial intelligence-based question processing method, device, equipment and storage medium. BACKGROUND

[0002] In a government scenario, the database (or answer base) of the question answering system only stores various standard answers (Answers, or government answers) related to government affairs. For a user input government question (Question), how to recall the standard answer corresponding to the government question from the database of the question answering system is very important. SUMMARY

[0003] The present disclosure provides an artificial intelligence-based question processing method, device, equipment and storage medium.

[0004] According to an aspect of the present disclosure, an artificial intelligence-based question processing method is provided, comprising:

[0005] obtaining a question to be processed;

[0006] determining a first correlation degree between the question to be processed and a plurality of standard answers in an answer base;

[0007] determining a target answer from the plurality of standard answers according to the first correlation degree of the plurality of standard answers;

[0008] replying to the question to be processed according to the target answer.

[0009] According to another aspect of the present disclosure, an artificial intelligence-based question processing device is provided, comprising:

[0010] a first obtaining module configured to obtain a question to be processed;

[0011] a first determining module configured to determine a first correlation degree between the question to be processed and a plurality of standard answers in an answer base;

[0012] a second determining module configured to determine a target answer from the plurality of standard answers according to the first correlation degree of the plurality of standard answers;

[0013] a reply module configured to reply to the question to be processed according to the target answer.

[0014] According to still another aspect of the present disclosure, an electronic device is provided, comprising:

[0015] at least one processor; and

[0016] a memory in communication with the at least one processor; wherein

[0017] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the artificial intelligence-based question processing method according to the aspect of the present disclosure.

[0018] According to still another aspect of the present disclosure, a non-transitory computer-readable storage medium of computer instructions is provided, the computer instructions being used to cause the computer to perform the artificial intelligence-based question processing method according to the aspect of the present disclosure.

[0019] According to still another aspect of the present disclosure, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the artificial intelligence-based question processing method according to the aspect of the present disclosure.

[0020] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0021] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them:

[0022] Figure 1 is a flowchart of the artificial intelligence-based question processing method provided by the first embodiment of the present disclosure;

[0023] Figure 2 is a flowchart of the artificial intelligence-based question processing method provided by the second embodiment of the present disclosure;

[0024] Figure 3 is a flowchart of the artificial intelligence-based question processing method provided by the third embodiment of the present disclosure;

[0025] Figure 4 is an input and output schematic diagram of the first language model provided by the embodiment of the present disclosure;

[0026] Figure 5 is a flowchart of the artificial intelligence-based question processing method provided by the fourth embodiment of the present disclosure;

[0027] Figure 6 is a flowchart of the artificial intelligence-based question processing method provided by the fifth embodiment of the present disclosure;

[0028] Figure 7 is an input and output schematic diagram of the third language model provided by the embodiment of the present disclosure;

[0029] Figure 8 A structural schematic diagram of the question processing apparatus based on artificial intelligence provided in Embodiment Six of the present disclosure is shown.

[0030] Figure 9 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0031] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding, which should be considered in a descriptive sense only. Thus, it will be apparent to one of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted from the following description.

[0032] At present, in the government affairs scene, only standard answers (Answers) related to government affairs are stored in the database of the question and answer system. For the user input government question (Question), the similarity between the government question and each standard answer in the database of the question and answer system is calculated to recall the standard answer corresponding to the government question.

[0033] For example, the Question can be vectorized to obtain the vectorized representation (Embedding) of the Question. The Embedding of the Question and the Embedding of each Answer in the database are calculated to obtain the similarity (or text similarity) between each Answer and the Question. Each Answer is sorted according to the value of the similarity from large to small, and the top N standard answers are recalled as the standard answers corresponding to the government question.

[0034] However, the text similarity between the Question and the Answer may not be high, resulting in inaccurate standard answers recalled.

[0035] For example, in a non-government affairs scene, assume that the user input question is "What should I eat to lose weight?", and the answer corresponding to the question is "broccoli, cucumber, tomato, chicken breast, corn, sweet potato, buckwheat noodles, and other low-calorie foods." From the above question and answer, it can be seen that there is only a correlation between the two, but they are not similar, that is, the text similarity between them is low. Therefore, it is quite difficult and challenging to directly use the question to answer the question.

[0036] Therefore, the present disclosure proposes a problem processing method and device based on artificial intelligence, an equipment and a storage medium.

[0037] The problem processing method and device based on artificial intelligence, the equipment and the storage medium of the embodiments of the present disclosure are described below with reference to the drawings.

[0038] Figure 1 The flowchart of the problem processing method based on artificial intelligence provided by the first embodiment of the present disclosure.

[0039] The embodiments of the present disclosure take the problem processing method based on artificial intelligence as an example and illustrate that the problem processing method based on artificial intelligence is configured in a problem processing device based on artificial intelligence. The problem processing device based on artificial intelligence can be applied to any electronic equipment, so that the electronic equipment can perform the problem processing function based on artificial intelligence.

[0040] The electronic equipment can be any device with computing power, such as a personal computer (PC), a mobile terminal, a server, etc. The mobile terminal can be a mobile phone, a tablet computer, a personal digital assistant, a wearable device, etc. with various operating systems, touch screens and / or display screens.

[0041] As shown in Figure 1 The problem processing method based on artificial intelligence can include the following steps:

[0042] Step S101, obtaining a problem to be processed.

[0043] In the embodiments of the present disclosure, the problem to be processed can be a problem input by a user. The input method includes but is not limited to touch input (such as sliding, clicking, etc.), keyboard input, voice input, etc.

[0044] Step S102, determining a first correlation degree between the problem to be processed and a plurality of standard answers in an answer library.

[0045] In the embodiments of the present disclosure, the answer library includes a plurality of standard answers.

[0046] In the embodiments of the present disclosure, the correlation degree between the problem to be processed and the plurality of standard answers in the answer library (denoted as the first correlation degree in the present disclosure) can be calculated.

[0047] Step S103, determining a target answer from the plurality of standard answers according to the first correlation degree of the plurality of standard answers.

[0048] In the embodiments of the present disclosure, the target answer can be determined from the plurality of standard answers according to the first correlation degree of the plurality of standard answers.

[0049] As an example, the multiple standard answers can be sorted according to the values of the corresponding first correlation degrees from large to small, and a set number of standard answers in the front of the sorting are selected as the target answers.

[0050] As another example, a standard answer with a first correlation degree higher than a set correlation degree threshold is taken as the target answer.

[0051] As yet another example, a standard answer with the highest first correlation degree is taken as the target answer.

[0052] In step S104, the target answer is used to reply to the problem to be processed.

[0053] In the embodiments of the present disclosure, the target answer can be used to reply to the problem to be processed.

[0054] As an example, the target answer can be used to automatically reply to the problem to be processed.

[0055] As another example, the target answer can also be provided to a customer service personnel, and the customer service personnel can manually reply to the problem to be processed according to the target answer, that is, the target answer can assist the customer service personnel to manually reply to the problem to be processed input by the user.

[0056] The problem processing method based on artificial intelligence in the embodiments of the present disclosure determines the first correlation degrees between the problem to be processed and the multiple standard answers in the answer library, determines the target answer from the multiple standard answers according to the first correlation degrees of the multiple standard answers, and replies to the problem to be processed according to the target answer. In this way, the reply answer (denoted as the target answer in the present disclosure) corresponding to the problem to be processed can be determined from the multiple standard answers based on the correlation degrees between the problem to be processed and the multiple standard answers in the answer library, and the accuracy of the determination of the target answer can be improved. In addition, in the scenario of manual reply by the customer service personnel, the determined target answer can be provided to the customer service personnel to assist the customer service personnel to quickly solve the problem of the user, which not only improves the service efficiency of the customer service personnel, but also improves the user experience.

[0057] It should be noted that in the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information are all carried out on the premise of obtaining the consent of the user, and all comply with the relevant legal regulations and do not violate public order and good customs.

[0058] In order to clearly illustrate how the first correlation degrees between the problem to be processed and the multiple standard answers in the answer library are determined in the above embodiments, the present disclosure further provides a problem processing method based on artificial intelligence.

[0059] Figure 2A flowchart of a question processing method based on artificial intelligence provided by Embodiment Two of the present disclosure.

[0060] As shown in Figure 2 The question processing method based on artificial intelligence can include the following steps:

[0061] Step S201, obtaining a question to be processed.

[0062] The explanation of step S201 can refer to the relevant description in any embodiment of the present disclosure, which will not be repeated here.

[0063] Step S202, using a first language model to predict an answer to the question to be processed to obtain at least one reference answer.

[0064] In an embodiment of the present disclosure, the first language model can be a pre-trained language model or a large language model.

[0065] In a possible implementation manner of the embodiment of the present disclosure, the manner of obtaining the reference answer is, for example, directly using the first language model to predict an answer to the question to be processed to obtain at least one reference answer output by the first language model.

[0066] In another possible implementation manner of the embodiment of the present disclosure, the manner of obtaining the reference answer is, for example:

[0067] 1. Obtain first prompt information, wherein the first prompt information is used to indicate task information to be processed by the first language model and a first number of reference answers output by the first language model.

[0068] In an embodiment of the present disclosure, the first prompt information can be manually input by a user or a relevant person, or can also be pre-generated, which is not limited by the present disclosure. The first prompt information can be used as prior information or task information to indicate information to be recognized by the first language model (or task information to be processed).

[0069] For example, taking 5 as the first number, the first prompt information (prompt) can be "please give 5 answers corresponding to the question", "output 5 answers corresponding to the question", and the like.

[0070] 2. Using the first language model to predict an answer to the question to be processed based on the first prompt information to obtain a first number of reference answers.

[0071] In summary, the first prompt information can be used as prior information or task information to indicate what information needs to be output by the first language model, which can improve the expression ability of the first language model and improve the prediction effect of the first language model.

[0072] Step S203, obtaining a text similarity between the reference answer and the plurality of standard answers.

[0073] In the embodiments of the present disclosure, the text similarity between the reference answer and the plurality of standard answers can be calculated respectively.

[0074] As an example, the reference answer can be vectorized to obtain a vectorized representation Embedding of the reference answer, and the Embeddings of the plurality of standard answers are obtained, the similarity between the Embedding of the reference answer and the Embeddings of the plurality of standard answers is calculated, and the similarity is taken as the text similarity between the reference answer and the plurality of standard answers.

[0075] Step S204, determining a first correlation degree between the to-be-processed question and the plurality of standard answers according to the text similarity between the reference answer and the plurality of standard answers.

[0076] In the embodiments of the present disclosure, for any one standard answer, the first correlation degree between the to-be-processed question and the standard answer can be determined according to the text similarity between the reference answer and the standard answer, wherein the first correlation degree is positively correlated with the text similarity, that is, the higher the text similarity, the higher the first correlation degree, and vice versa, the lower the text similarity, the lower the first correlation degree.

[0077] Step S205, determining a target answer from the plurality of standard answers according to the first correlation degrees of the plurality of standard answers.

[0078] Step S206, replying to the to-be-processed question according to the target answer.

[0079] The explanation and description of steps S205 to S206 can be referred to the related description in any embodiment of the present disclosure, which will not be repeated here.

[0080] The question processing method based on artificial intelligence in the embodiments of the present disclosure can improve the accuracy of recalling the target answer by predicting the reference answer corresponding to the to-be-processed question, and recalling the target answer corresponding to the to-be-processed question based on the text similarity between the reference answer and the standard answer. That is, in the present application, it is considered that the question and the answer may not be similar, and the similarity between the answers can be high, therefore, the target answer corresponding to the to-be-processed question can be recalled based on the text similarity between the reference answer and the standard answer, and the accuracy of recalling the target answer can be improved.

[0081] In order to clearly illustrate how the first correlation degree between the to-be-processed question and the plurality of standard answers is determined in the above embodiments, the present disclosure further proposes a question processing method based on artificial intelligence.

[0082] Figure 3This is a flowchart illustrating the problem-solving method based on artificial intelligence provided in Embodiment 3 of this disclosure.

[0083] like Figure 3 As shown, this AI-based problem-solving method may include the following steps:

[0084] Step S301: Obtain the problem to be processed.

[0085] The explanation of step S301 can be found in the relevant description in any embodiment of this disclosure, and will not be repeated here.

[0086] Step S302: Use a first language model to predict the answer to the problem to be processed, so as to obtain multiple reference answers.

[0087] In this embodiment of the disclosure, a first language model can be used to predict the answer to the problem to be processed, so as to obtain multiple reference answers output by the first language model.

[0088] For example, let's take a non-governmental scenario as an example. Suppose the user inputs the problem to be solved as follows: Figure 4 The first prompt in area 41, "What foods are suitable to eat during weight loss?", is... Figure 4 The "Please provide 5 answers" option shown in area 42 allows you to input both the question to be processed and the initial prompt information into the first language model simultaneously. In this case, the first language model can output something like... Figure 4 The five reference answers shown in section 43 are as follows:

[0089] 1. Vegetables: Vegetables are rich in fiber and vitamins, which can help you feel full and reduce food intake. In particular, non-starchy vegetables, such as spinach, cauliflower, carrots, and broccoli, are low in calories and rich in vitamins and minerals.

[0090] 2. Fruits: Fruits are also suitable foods during weight loss. They are rich in dietary fiber and water, which can help satisfy cravings and provide the body with necessary vitamins and minerals. However, it is important to control intake, as some fruits are high in sugar.

[0091] 3. Lean meat: Lean meat is rich in protein, which can help you maintain muscle mass and speed up metabolism, thus aiding in weight loss. Choose lean meats such as chicken, turkey, and fish, and avoid processed meat products.

[0092] 4. Legumes: Legumes are a type of food rich in protein, fiber, and low-GI (glycemic index) carbohydrates, which can help control hunger and provide sustained energy. Good choices include tofu, black beans, and red beans.

[0093] 5. Whole grains: Whole grains are packed with dietary fiber, low GI carbohydrates, and other nutrients that can help you feel full and provide sustained energy. Choose whole grains such as oatmeal, whole wheat bread, brown rice, and so on.

[0094] In step S303, for any one of the standard answers in the answer library, text similarities between the plurality of reference answers and the any one of the standard answers are respectively obtained.

[0095] In the embodiments of the present disclosure, for any one of the standard answers in the answer library, text similarities between the plurality of reference answers and the any one of the standard answers can be respectively calculated.

[0096] In step S304, according to the text similarities between the plurality of reference answers and the plurality of standard answers, a first correlation degree between the to-be-processed question and the plurality of standard answers is determined.

[0097] In the embodiments of the present disclosure, for any one of the standard answers in the answer library, according to the text similarities between the plurality of reference answers and the any one of the standard answers, a first correlation degree between the to-be-processed question and the any one of the standard answers can be determined.

[0098] In a possible implementation manner of the embodiments of the present disclosure, when there is one first correlation degree between the to-be-processed question and each of the standard answers, the first correlation degree between the to-be-processed question and one of the standard answers is calculated in the following manner, for example: fusing (such as weighted summation) the text similarities between the plurality of reference answers and the one of the standard answers to obtain a first fused similarity, and determining the first correlation degree between the to-be-processed question and the one of the standard answers according to the first fused similarity.

[0099] In the embodiments of the present disclosure, the first correlation degree and the first fused similarity are in a positive correlation relationship.

[0100] In another possible implementation manner of the embodiments of the present disclosure, when there is one first correlation degree between the to-be-processed question and each of the standard answers, the first correlation degree between the to-be-processed question and one of the standard answers is calculated in the following manner, for example: obtaining a statistical value (denoted as a first statistical value in the present disclosure, wherein the first statistical value includes at least one of a maximum value, a minimum value, a mean value, and a median value) of the text similarities between the plurality of reference answers and the one of the standard answers, and determining the first correlation degree between the to-be-processed question and the one of the standard answers according to the first statistical value.

[0101] In the embodiments of the present disclosure, the first correlation degree and the first statistical value are in a positive correlation relationship.

[0102] Therefore, the first correlation degree between the to-be-processed question and the standard answer can be determined according to different manners, and the flexibility and applicability of the method can be improved.

[0103] In step S305, the target answer is determined from the multiple standard answers according to the first correlation degrees of the multiple standard answers.

[0104] In a possible implementation of the embodiments of the present disclosure, when there is one first correlation degree between the to-be-processed question and each standard answer, the target answer can be determined from the multiple standard answers according to the first correlation degree corresponding to each standard answer.

[0105] As an example, the multiple standard answers can be sorted according to the values of the corresponding first correlation degrees from large to small, and a set number of standard answers in the front of the sorting are selected as the target answers.

[0106] As another example, the standard answers whose first correlation degrees are higher than a set correlation degree threshold value are selected as the target answers.

[0107] As yet another example, the standard answer with the highest first correlation degree is selected as the target answer.

[0108] In another possible implementation of the embodiments of the present disclosure, when there are multiple first correlation degrees between the to-be-processed question and each standard answer, each first correlation degree is determined according to the text similarity between a reference answer and a standard answer, and the target answer is determined in the following manner, for example:

[0109] 1. For any one of the multiple reference answers, an answer set is determined from the multiple standard answers based on the first correlation degrees determined according to the text similarity between the reference answer and the multiple standard answers; wherein the answer set includes at least one standard answer, and the first correlation degree determined according to the text similarity between the at least one standard answer and the reference answer is higher than a set correlation degree threshold value.

[0110] That is, for each reference answer, an answer set can be selected for the reference answer, wherein the text similarity between each standard answer in the answer set and the reference answer is higher than a set similarity threshold value, so that the first correlation degrees between the to-be-processed question and each standard answer in the answer set determined according to the text similarity between the reference answer and each standard answer in the answer set are higher than a set correlation degree threshold value.

[0111] 2. The target answer is determined from the answer sets corresponding to the multiple reference answers according to the occurrence frequencies of the standard answers in the answer sets corresponding to the multiple reference answers.

[0112] For example, when the number of reference answers is M, the occurrence frequencies of the standard answers in the M answer sets can be counted, and the target answer with a relatively high occurrence frequency can be determined from the M answer sets.

[0113] For example, assuming that M is 3, the 3 answer sets are set 1, set 2 and set 3, set 1 includes standard answer 1, standard answer 2, standard answer 3 and standard answer 4, set 2 includes standard answer 2, standard answer 3, standard answer 4 and standard answer 5, and set 3 includes standard answer 2, standard answer 3 and standard answer 5, the occurrence number of standard answer 1 is 1, the occurrence number of standard answer 2 is 3, the occurrence number of standard answer 3 is 3, the occurrence number of standard answer 4 is 2, and the occurrence number of standard answer 5 is 2, thus, the target answer can be standard answer 2 and standard answer 3.

[0114] Therefore, the target answer can be determined based on the standard answer similar to the multiple reference answers, or the target answer can be determined based on the standard answer similar to the majority of the reference answers, and the rationality and accuracy of the target answer determination can be improved.

[0115] In step S306, the to-be-processed question is replied according to the target answer.

[0116] The explanation of steps S305 to S306 can refer to the related description in any embodiment of the present disclosure, which will not be repeated here.

[0117] The question processing method based on artificial intelligence provided in the embodiment of the present disclosure can improve the accuracy and reliability of the determination result by determining the first correlation degree between the to-be-processed question and the multiple standard answers according to the text similarity between all the reference answers and the multiple standard answers.

[0118] In order to clearly illustrate how the first correlation degree between the to-be-processed question and the multiple standard answers in the answer library is determined in any embodiment of the present disclosure, the present disclosure further provides a question processing method based on artificial intelligence.

[0119] Figure 5 A flowchart of the question processing method based on artificial intelligence provided in the fourth embodiment of the present disclosure is shown.

[0120] As Figure 5 shown, the question processing method based on artificial intelligence can include the following steps:

[0121] In step S501, a to-be-processed question and multiple sentence groups are obtained, each of which includes a standard answer in an answer library and at least one reference question, wherein the at least one reference question is obtained by using a second language model to predict a question for the standard answer.

[0122] The obtaining method of the to-be-processed question can refer to the related description in any of the above embodiments, which will not be repeated here.

[0123] In the embodiments of the present disclosure, the second language model is also a pre-trained language model or a large language model. The second language model can be the same language model as the first language model, or can be a different language model, and the embodiments of the present disclosure do not limit this.

[0124] In a possible implementation manner of the embodiments of the present disclosure, the obtaining manner of each sentence group is, for example:

[0125] 1. Directly using the second language model to perform question prediction on a standard answer in the answer library to obtain at least one reference question corresponding to the standard answer.

[0126] 2. Generating a sentence group according to the standard answer and the at least one reference question.

[0127] In a possible implementation manner of the embodiments of the present disclosure, the generation manner of each sentence group is, for example:

[0128] 1. Obtaining second prompt information, wherein the second prompt information is used to indicate task information to be processed by the second language model and a second quantity of reference questions output by the second language model.

[0129] In the embodiments of the present disclosure, the second prompt information can be manually input by a user or a related person, or can be pre-generated, and the present disclosure does not limit this. The second prompt information can be used as prior information or task information to indicate information to be recognized by the second language model (or task information to be processed).

[0130] In the embodiments of the present disclosure, the size relationship between the first quantity and the second quantity is not limited, for example, the first quantity can be equal to the second quantity, or the first quantity can be greater than the second quantity, or the first quantity can be less than the second quantity.

[0131] For example, taking the second quantity as 3 for example, the second prompt information (prompt) can be “please give 3 questions corresponding to the answer”, “output 3 questions corresponding to the answer”, and the like.

[0132] 2. Selecting a standard answer from the answer library.

[0133] For example, a standard answer can be selected from the answer library in sequence, or a standard answer can be randomly selected from the answer library.

[0134] 3. Using the second language model to perform question prediction on the standard answer based on the second prompt information to obtain a second quantity of reference questions corresponding to the standard answer.

[0135] 4. generating a sentence group according to the second number of reference questions and a standard answer.

[0136] In summary, the second prompt information can be used as prior information or task information to indicate what information the second language model needs to output, thereby improving the expression ability of the second language model and the prediction effect of the second language model.

[0137] In step S502, text similarities between the to-be-processed question and each reference question in the plurality of sentence groups are obtained.

[0138] In the embodiments of the present disclosure, for any one of the plurality of sentence groups, the text similarities between the to-be-processed question and each reference question in the sentence group can be calculated respectively.

[0139] As an example, the to-be-processed question can be vectorized to obtain a vectorized representation Embedding of the to-be-processed question, and the Embeddings of each reference question in the sentence group are obtained, and the similarity between the Embedding of the to-be-processed question and the Embeddings of each reference question in the sentence group is calculated and used as the text similarity between the to-be-processed question and each reference question in the sentence group.

[0140] In step S503, a first correlation degree between the to-be-processed question and the standard answer in the plurality of sentence groups is determined according to the text similarities between the to-be-processed question and each reference question in the plurality of sentence groups.

[0141] In the embodiments of the present disclosure, for any one of the plurality of sentence groups, the first correlation degree between the to-be-processed question and the standard answer in the sentence group can be determined according to the text similarities between the to-be-processed question and each reference question in the sentence group. The first correlation degree is positively correlated with the text similarity.

[0142] In a possible implementation manner of the embodiments of the present disclosure, the text similarities between the to-be-processed question and each reference question in the above-mentioned sentence group can be fused (such as weighted summation) to obtain a second fused similarity, and the first correlation degree between the to-be-processed question and the standard answer in the sentence group is determined according to the second fused similarity.

[0143] The first correlation degree is positively correlated with the second fused similarity.

[0144] In another possible implementation manner of the embodiments of the present disclosure, a statistical value (denoted as a second statistical value in the present disclosure, wherein the second statistical value includes at least one of a maximum value, a minimum value, a mean value and a median value) of the text similarities between the to-be-processed question and each reference question in the sentence group is obtained, and the first correlation degree between the to-be-processed question and the standard answer in the sentence group is determined according to the second statistical value.

[0145] The first correlation degree is positively correlated with the second statistical value.

[0146] Therefore, the first correlation degree between the to-be-processed question and the standard answers in each sentence group can be determined based on different manners, and the flexibility and applicability of the method can be improved.

[0147] In step S504, a target answer is determined from the plurality of standard answers according to the first correlation degrees of the plurality of standard answers.

[0148] In step S505, the to-be-processed question is replied according to the target answer.

[0149] The explanation and description of steps S504 to S505 can be referred to the related description in any embodiment of the present disclosure, which will not be repeated here.

[0150] The question processing method based on artificial intelligence in the embodiments of the present disclosure can improve the accuracy of recalling the target answer by predicting the reference questions corresponding to each standard answer in the answer library and recalling the target answer corresponding to the to-be-processed question based on the text similarity between the reference question and the to-be-processed question. That is, in the present application, it is considered that the question and the answer may not be similar, but the similarity between the questions may be high. Therefore, the target answer corresponding to the to-be-processed question can be recalled based on the text similarity between the reference question and the to-be-processed question, and the accuracy of recalling the target answer can be improved.

[0151] In order to clearly illustrate how the to-be-processed question is replied according to the target answer in any embodiment of the present disclosure, the present disclosure further provides a question processing method based on artificial intelligence.

[0152] Figure 6 A flowchart of the question processing method based on artificial intelligence provided in the fifth embodiment of the present disclosure.

[0153] As shown in the flowchart, the question processing method based on artificial intelligence can include the following steps: Figure 6

[0154] In step S601, a to-be-processed question is obtained.

[0155] In step S602, a first correlation degree between the to-be-processed question and a plurality of standard answers in an answer library is determined.

[0156] In step S603, a target answer is determined from the plurality of standard answers according to the first correlation degrees of the plurality of standard answers, wherein the target answer is multiple.

[0157] The explanation and description of steps S601 to S603 can be referred to the related description in any embodiment of the present disclosure, which will not be repeated here.​

[0158] As an example, the multiple standard answers can be sorted according to the values of the corresponding first correlation degrees from large to small, and a set number of standard answers in the front of the sorting are selected as the target answers.

[0159] The set number is greater than 1.

[0160] As another example, a standard answer whose first correlation degree is higher than a set correlation degree threshold can be selected as the target answer.

[0161] In step S604, a second correlation degree between the multiple target answers and the problem to be processed is obtained.

[0162] In the embodiments of the present disclosure, the second correlation degrees between the multiple target answers and the problem to be processed can be calculated respectively.

[0163] As a possible implementation manner, the calculation manner of the second correlation degree can be, for example:

[0164] 1. For any one target answer, user feedback information listened to by the target answer in at least one dimension is obtained, wherein each dimension corresponds to one user interaction behavior.

[0165] The user interaction behavior includes but is not limited to adoption, clicking, browsing, commenting, following, collecting, sharing, etc. The user feedback information includes but is not limited to the number of times the target answer is adopted, the number of clicks, the browsing time, the number of comments, the number of followings, the number of collections, the number of shares, etc.

[0166] 2. According to the user feedback information in at least one dimension, a score of the target answer in at least one dimension is determined.

[0167] The score in each dimension is positively correlated with the user feedback information in the corresponding dimension, for example, the more times the target answer is adopted, the higher the score in the corresponding dimension.

[0168] 3. According to the score of the target answer in at least one dimension, a second correlation degree between the target answer and the problem to be processed is determined.

[0169] The second correlation degree is positively correlated with the score in at least one dimension.

[0170] As an example, the scores of the target answer in at least one dimension can be weighted and summed to obtain the second correlation degree between the target answer and the problem to be processed.

[0171] As another example, the scores of the target answer in at least one dimension can be accumulated, averaged, etc. to obtain the second correlation degree between the target answer and the problem to be processed.

[0172] Therefore, based on the user feedback information listened to by the target answer in at least one dimension, the second correlation degree between the target answer and the to-be-processed question is determined, which can realize recommending the popular target answer to the user, so that the target answer replied to the user is more in line with the actual needs of the user, and the user experience is improved.

[0173] As another possible implementation manner, the calculation manner of the second correlation degree may be, for example:

[0174] 1. Obtain third prompt information, wherein the third prompt information is used to prompt task information to be processed by the third language model.

[0175] In the embodiments of the present disclosure, the third prompt information can be manually input by the user or relevant personnel, or can also be pre-generated, and the present disclosure does not make any limitation thereon. The third prompt information can be used as prior information or task information to indicate information to be recognized by the third language model (or task information to be processed).

[0176] In the embodiments of the present disclosure, the third language model is also a pre-trained language model or a large language model. The third language model and the first language model can be the same language model, or can also be different language models, and the embodiments of the present disclosure do not make any limitation thereon.

[0177] 2. Using the third language model, the correlation degree between the to-be-processed question and the plurality of target answers is predicted based on the third prompt information to obtain a correlation degree score between the to-be-processed question and the plurality of target answers.

[0178] The correlation degree score is used to indicate the correlation degree between the to-be-processed question and the target answer.

[0179] 3. According to the correlation degree scores of the plurality of target answers, the second correlation degree between the plurality of target answers and the to-be-processed question is determined.

[0180] The second correlation degree is positively correlated with the correlation degree score, that is, the higher the correlation degree score between the target answer and the to-be-processed question, the higher the second correlation degree between the target answer and the to-be-processed question.

[0181] As an example, the to-be-processed question is Figure 7 The number of target answers is 3, and the 3 target answers are Figure 7 The third prompt information is Figure 7As shown in the "Please rank the three answers and the question according to the relevance and give the relevance score from 1-10" in the middle area 73, the output of the third language model can be as follows Figure 7 As shown in the middle area 74, the area 74 includes the relevance score between the three target answers output by the third language model and the question to be processed.

[0182] In this way, the trained language model is used to predict the relevance between the question to be processed and the target answer, which can improve the accuracy of the prediction result.

[0183] As another possible implementation, the second correlation degree can be calculated in the following manner, for example:

[0184] 1. Calculate the semantic similarity between the question to be processed and the plurality of target answers, respectively.

[0185] 2. Determine the second correlation degree between the plurality of target answers and the question to be processed based on the semantic similarity of the plurality of target answers.

[0186] The second correlation degree is positively correlated with the semantic similarity, that is, the higher the semantic similarity between the target answer and the question to be processed, the higher the second correlation degree between the target answer and the question to be processed.

[0187] In this way, the second correlation degree between the question to be processed and the target answer is determined based on the semantic similarity between the question to be processed and the target answer, which can improve the accuracy of the second correlation degree calculation. Moreover, the second correlation degree between the target answer and the question to be processed can be calculated in different ways, which can improve the flexibility and applicability of the method.

[0188] In step S605, the plurality of target answers are ranked based on the second correlation degree of the plurality of target answers, to obtain a ranking sequence.

[0189] In the embodiments of the present disclosure, the plurality of target answers can be ranked from large to small according to the values of the corresponding second correlation degrees, to obtain the ranking sequence.

[0190] In step S606, the question to be processed is replied according to the target answer ranked first in the ranking sequence.

[0191] In the embodiments of the present disclosure, the question to be processed can be replied according to the target answer ranked first in the ranking sequence.

[0192] As an example, the question to be processed can be automatically replied according to the target answer ranked first in the ranking sequence.

[0193] As another example, the target answer ranked first in the ranking sequence can also be provided to the customer service personnel, and the customer service personnel manually replies to the to-be-processed question according to the target answer ranked first, that is, the role of the target answer is to assist the customer service personnel in manually replying to the to-be-processed question input by the user.

[0194] The question processing method based on artificial intelligence in the embodiments of the present disclosure can determine the target answer for replying to the to-be-processed question from the plurality of target answers based on the second correlation degree between the plurality of target answers and the to-be-processed question when the target answer is multiple, which can improve the accuracy of the reply to the to-be-processed question and improve the user experience.

[0195] In any one of the embodiments of the present disclosure, the first language model, the second language model, and the third language model are all large language models, and in the present disclosure, a large language model can be used to improve the recall ability of the answer, for example, the recall ability of the answer can be improved in the following two aspects.

[0196] First, improve the online recall ability.

[0197] A) use a large language model to predict answers to the to-be-processed question Question to obtain a plurality of reference answers Answer;

[0198] B) vectorize the plurality of reference answers to obtain Embedding of the plurality of reference answers, and perform similarity calculation on the Embedding of the plurality of reference answers and the Embedding of each standard answer in the answer library to obtain the target answer in the answer library.

[0199] This scheme can optimize the problem of low accuracy of answer recall caused by directly using Question to recall Answer in the answer library in related technologies.

[0200] For example, assume that the to-be-processed question is: What should I eat to lose weight? The target answer is: broccoli, cucumber, tomato, chicken breast, corn, sweet potato, buckwheat noodles, and other low-calorie foods. According to the above example, the text similarity between the to-be-processed question and the target answer is low, and it is quite difficult and challenging to directly use the to-be-processed question to recall the target answer in the answer library.

[0201] In the present application, a large language model can be used to predict a plurality of reference answers corresponding to the to-be-processed question, for example, the reference answers can be:

[0202] Reference answer 1: The following foods can be selected during weight loss:

[0203] Breakfast: milk, freshly squeezed juice, eggs, boiled cabbage, broccoli, avocado, boiled cauliflower, etc.

[0204] Lunch: chicken breast, beef, fruit and vegetable salad.

[0205] Dinner: low-calorie fruits and vegetables such as cucumbers and tomatoes.

[0206] Reference answer 2: high-protein, low-calorie foods such as lean meat, eggs, tofu, fish, chicken breast, leafy vegetables, fruits, nuts, etc.

[0207] From the multiple reference answers output by the large language model, it is obvious that it is easier to recall similar target answers to the reference answers.

[0208] Second aspect: expand the reference questions corresponding to each standard answer in the offline answer library.

[0209] The offline answer library can store each standard answer and the Embedding corresponding to the standard answer. The reference questions corresponding to these standard answers can be predicted by the large language model. Based on the text similarity between the reference questions and the problems to be processed, the target answers in the answer library are recalled.

[0210] As an example, for the standard answer "food to eat during weight loss: broccoli, cucumber, tomato, chicken breast, corn, sweet potato, buckwheat noodles, etc. low-calorie food" in the answer library, a large language model is used to predict multiple reference questions corresponding to the standard answer. For example, a large language model 1 can be used to predict the following 5 reference questions:

[0211] Reference question 1: What meat can be eaten during weight loss?

[0212] Reference question 2: What vegetables can be eaten during weight loss?

[0213] Reference question 3: How should you control your diet during weight loss?

[0214] Reference question 4: What dietary principles should be followed during weight loss?

[0215] Reference question 5: What main dishes can be eaten during weight loss?

[0216] For another example, a large language model 2 can be used to predict the following 5 reference questions:

[0217] Reference question 1: What are some low-calorie foods?

[0218] Reference question 2: What are the benefits of broccoli, cucumber, and sweet potato during weight loss?

[0219] Reference question 3: Is chicken breast a good weight loss food? Why?

[0220] Reference question 4: Can corn and buckwheat noodles be used as main dishes in weight loss diets?

[0221] Reference question 5: What other foods are suitable for weight loss besides these foods?

[0222] In the present disclosure, the five reference questions output by the large language model and a standard answer can form a sentence group, so that the target sentence group can be determined from each sentence group based on the text similarity between the to-be-processed question and the reference questions in each sentence group, wherein the target sentence group contains a reference question matching or similar to the to-be-processed question, and the standard answer in the target sentence group is taken as the target answer.

[0223] That is, in the present application, the multiple reference questions output by the large language model can be stored offline in the answer library to improve the recall rate of the target answer.

[0224] Corresponding to the question processing method based on artificial intelligence provided by the above Figures 1 to 6 The embodiment provides a question processing method based on artificial intelligence. Figures 1 to 6 Corresponding to the question processing method based on artificial intelligence provided by the above

[0225] Figure 8 The structure of the question processing device based on artificial intelligence provided in the sixth embodiment of the present disclosure is shown in the figure.

[0226] As Figure 8 shown, the question processing device based on artificial intelligence 800 can include a first acquisition module 801, a first determination module 802, a second determination module 803, and a reply module 804.

[0227] The first acquisition module 801 is configured to acquire a to-be-processed question.

[0228] The first determination module 802 is configured to determine a first correlation degree between the to-be-processed question and multiple standard answers in an answer library.

[0229] The second determination module 803 is configured to determine a target answer from the multiple standard answers according to the first correlation degrees of the multiple standard answers.

[0230] The reply module 804 is configured to reply to the to-be-processed question according to the target answer.

[0231] In a possible implementation manner of the embodiment of the present disclosure, the first determining module 802 is configured to: perform answer prediction on the to-be-processed question by using the first language model to obtain at least one reference answer; obtain text similarity between the reference answer and the plurality of standard answers; and determine the first correlation degree between the to-be-processed question and the plurality of standard answers according to the text similarity between the reference answer and the plurality of standard answers.

[0232] In a possible implementation manner of the embodiment of the present disclosure, there is one first correlation degree between the to-be-processed question and each standard answer when the reference answer is a plurality of reference answers; the first determining module 802 is configured to: for any one of the plurality of standard answers, fuse the text similarity between the plurality of reference answers and the any one of the standard answers to obtain a first fused similarity, and determine the first correlation degree between the to-be-processed question and the any one of the standard answers according to the first fused similarity; or for any one of the plurality of standard answers, obtain a first statistical value of the text similarity between the plurality of reference answers and the any one of the standard answers, and determine the first correlation degree between the to-be-processed question and the any one of the standard answers according to the first statistical value; wherein the first statistical value includes at least one of a maximum value, a minimum value, a mean value and a median value.

[0233] In a possible implementation manner of the embodiment of the present disclosure, there are a plurality of first correlation degrees between the to-be-processed question and each standard answer when the reference answer is a plurality of reference answers, and the first correlation degree is determined according to the text similarity between one reference answer and one standard answer; the second determining module 803 is configured to: for any one of the plurality of reference answers, determine an answer set from the plurality of standard answers based on the first correlation degrees determined according to the text similarity between the any one of the reference answers and the plurality of standard answers; wherein the answer set includes at least one standard answer, and the first correlation degree determined according to the text similarity between the at least one standard answer and the any one of the reference answers is higher than a correlation degree threshold; and determine a target answer from the answer sets corresponding to the plurality of reference answers according to the number of occurrences of each standard answer in the answer sets corresponding to the plurality of reference answers.

[0234] In a possible implementation manner of the embodiment of the present disclosure, the first determining module 802 is configured to: obtain first prompt information, wherein the first prompt information is used to indicate task information to be processed by the first language model and a first number of reference answers output by the first language model; and perform answer prediction on the to-be-processed question based on the first prompt information by using the first language model to obtain the first number of reference answers.

[0235] In a possible implementation manner of the embodiment of the present disclosure, the first determining module 802 is configured to: obtain a plurality of sentence groups, wherein each of the sentence groups comprises one standard answer in the answer bank and at least one reference question, and the at least one reference question is obtained by performing question prediction on one standard answer by using the second language model; obtain text similarity between the to-be-processed question and each reference question in the plurality of sentence groups; and determine a first correlation degree between the to-be-processed question and the standard answer in the plurality of sentence groups according to the text similarity between the to-be-processed question and each reference question in the plurality of sentence groups.

[0236] In a possible implementation manner of the embodiment of the present disclosure, the sentence group is obtained by using the following module:

[0237] The second obtaining module is configured to obtain second prompt information, wherein the second prompt information is used to indicate task information to be processed by the second language model and a second quantity of reference questions output by the second language model.

[0238] The selecting module is configured to select one standard answer from the answer bank.

[0239] The predicting module is configured to perform question prediction on one standard answer based on the second prompt information by using the second language model, to obtain the second quantity of reference questions.

[0240] The generating module is configured to generate one sentence group according to the second quantity of reference questions and one standard answer.

[0241] In a possible implementation manner of the embodiment of the present disclosure, the first determining module 802 is configured to: for any sentence group, fuse the text similarity between the to-be-processed question and each reference question in the any sentence group to obtain a second fused similarity, and determine the first correlation degree between the to-be-processed question and the standard answer in the any sentence group according to the second fused similarity; or for any sentence group, obtain a second statistical value of the text similarity between the to-be-processed question and each reference question in the any sentence group, and determine the first correlation degree between the to-be-processed question and the standard answer in the any sentence group according to the second statistical value; wherein the second statistical value comprises at least one of a maximum value, a minimum value, a mean value and a median value.

[0242] In a possible implementation manner of the embodiment of the present disclosure, the target answer is a plurality of, and the replying module 804 is configured to: obtain a second correlation degree between the plurality of target answers and the to-be-processed question; sort the plurality of target answers based on the second correlation degree, to obtain a sorting sequence; and reply to the to-be-processed question according to a target answer in front of the sorting sequence.

[0243] In a possible implementation of the present disclosure, the reply module 803 is configured to: obtain user feedback information listened to by any target answer in at least one dimension for any target answer, wherein each dimension corresponds to a user interaction behavior; determine a score of any target answer in at least one dimension according to the user feedback information in at least one dimension; and determine a second correlation degree between any target answer and the to-be-processed question according to the score of any target answer in at least one dimension.

[0244] In a possible implementation of the present disclosure, the reply module 803 is configured to: obtain third prompt information, wherein the third prompt information is used to prompt task information to be processed by the third language model; and perform correlation degree prediction on the to-be-processed question and the plurality of target answers based on the third prompt information by using the third language model, to obtain a correlation degree score between the to-be-processed question and the plurality of target answers; and determine a second correlation degree between the plurality of target answers and the to-be-processed question according to the correlation degree scores of the plurality of target answers.

[0245] In a possible implementation of the present disclosure, the reply module 803 is configured to: obtain semantic similarity between the to-be-processed question and the plurality of target answers; and determine a second correlation degree between the plurality of target answers and the to-be-processed question based on the semantic similarity of the plurality of target answers.

[0246] The question processing apparatus based on artificial intelligence in the present disclosure determines a first correlation degree between a to-be-processed question and a plurality of standard answers in an answer library, determines a target answer from the plurality of standard answers according to the first correlation degree of the plurality of standard answers, and replies to the to-be-processed question according to the target answer. In this way, the corresponding reply answer (denoted as a target answer in the present disclosure) of the to-be-processed question can be determined from the plurality of standard answers based on the correlation degree between the to-be-processed question and the plurality of standard answers in the answer library, which can improve the accuracy of target answer determination. In addition, in the scenario of manual reply of customer service personnel, the determined target answer can be provided to the customer service personnel to assist the customer service personnel in quickly solving user problems, which can not only improve the service efficiency of the customer service personnel, but also improve the user experience.

[0247] To implement the above-mentioned embodiments, the present disclosure further provides an electronic device, which can include at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the question processing method based on artificial intelligence proposed in any of the embodiments of the present disclosure.

[0248] To achieve the above-mentioned embodiments, the present disclosure further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to make a computer execute the question processing method based on artificial intelligence proposed in any one of the above-mentioned embodiments of the present disclosure.

[0249] To achieve the above-mentioned embodiments, the present disclosure further provides a computer program product, which comprises a computer program, and the computer program realizes the question processing method based on artificial intelligence proposed in any one of the above-mentioned embodiments of the present disclosure when executed by a processor.

[0250] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium and a computer program product.

[0251] Figure 9 A schematic block diagram of an example electronic device that can be used to implement embodiments of the present disclosure is shown.

[0252] The electronic device can include a server, a client in the above-mentioned embodiments. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present disclosure described and / or claimed in this document.

[0253] As shown in Figure 9 The device 900 includes a computing unit 901 that can perform various appropriate actions and processes in accordance with a computer program stored in a ROM (Read-Only Memory) 902 or a computer program loaded into a RAM (Random Access Memory) 903 from a storage unit 909. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other through a bus 904. An I / O (Input / Output) interface 905 is also connected to the bus 904.

[0254] A plurality of components in the device 900 are connected to the I / O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a magnetic disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the device 900 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0255] The computing unit 901 can be various general and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 901 performs various methods and processes described above, such as the above-described AI-based question processing method. For example, in some embodiments, the above-described AI-based question processing method can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded to the RAM 903 and executed by the computing unit 901, one or more steps of the above-described AI-based question processing method can be performed. Alternatively, in other embodiments, the computing unit 901 can be configured to perform the above-described AI-based question processing method by any other appropriate means, such as by means of firmware.

[0256] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a Field Programmable Gate Array (FPGA), an Application-Specific Integrated Circuit (ASIC), an Application Specific Standard Product (ASSP), a System on a Chip (SOC), a Complex Programmable Logic Device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0257] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general or special purpose computer, such that the program code, when executed by the processor or controller, causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be implemented in a wholly in machine language, in partially in machine language, in partially in a high level language, and other combinations thereof. The program code can execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0258] In the context of this disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include a linearly-programmed electronic storage, a portable computer diskette, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory), or flash memory, an optical fiber, a CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0259] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0260] The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, and a blockchain network.

[0261] The computer system can include clients and servers. This relationship can be between a client and a server that are typically distant from each other and typically interact through a communication network. The relationship between the client and the server is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, and solves the defects of large management difficulty and weak business scalability in traditional physical hosts and VPS services (Virtual Private Server, virtual private server).

[0262] It should be noted that artificial intelligence is a discipline that studies enabling computers to simulate some thinking processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.) of people, and has both hardware and software technologies. Artificial intelligence hardware technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing, etc.; artificial intelligence software technology mainly includes computer vision technology, speech recognition technology, natural language processing technology, and machine learning / deep learning, big data processing technology, knowledge graph technology, etc. several major directions.

[0263] According to the technical scheme of the embodiment of the present disclosure, the first correlation degree between the to-be-processed question and the plurality of standard answers in the answer library is determined; the target answer is determined from the plurality of standard answers according to the first correlation degree of the plurality of standard answers; and the to-be-processed question is replied according to the target answer. Thus, the reply answer (denoted as the target answer in the present disclosure) corresponding to the to-be-processed question can be determined from the plurality of standard answers based on the correlation degree between the to-be-processed question and the plurality of standard answers in the answer library, which can improve the accuracy of the target answer determination. In addition, in the customer service personnel manual reply scene, the determined target answer can be provided to the customer service personnel to assist the customer service personnel to quickly solve the user problem, which not only can improve the service efficiency of the customer service personnel, but also can improve the user experience.

[0264] It should be understood that the various forms of the flow shown above can be used to reorder, add or delete steps. For example, the steps described in the present disclosure can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical scheme of the present disclosure can be achieved, which is not limited herein.

[0265] The above detailed description does not limit the scope of the disclosure. Various modifications, combinations, sub-combinations and alternatives can be made to the detailed description. Any modification, equivalent replacement and improvement etc. made within the spirit and principle of the disclosure shall be included in the scope of the disclosure.

Claims

1. An artificial intelligence-based question processing method, comprising: obtaining a to-be-processed question; determining a first correlation degree between the to-be-processed question and a plurality of standard answers in an answer library; determining a target answer from the plurality of standard answers according to the first correlation degrees of the plurality of standard answers; replying to the to-be-processed question according to the target answer; wherein the determining the first correlation degree between the to-be-processed question and the plurality of standard answers in the answer library comprises: performing answer prediction on the to-be-processed question by using a first language model to obtain at least one reference answer; obtaining a text similarity between the reference answer and the plurality of standard answers; determining the first correlation degree between the to-be-processed question and the plurality of standard answers according to the text similarity between the reference answer and the plurality of standard answers.

2. The method of claim 1, wherein, The reference answer is a plurality of, and there is a first correlation degree between the to-be-processed question and each standard answer. The determining the first correlation degree between the to-be-processed question and the plurality of standard answers according to the text similarity between the reference answer and the plurality of standard answers comprises: for any one of the plurality of standard answers, fusing the text similarities between the plurality of reference answers and the any one of the standard answers to obtain a first fused similarity, and determining the first correlation degree between the to-be-processed question and the any one of the standard answers according to the first fused similarity; or, for any one of the plurality of standard answers, obtaining a first statistical value of the text similarities between the plurality of reference answers and the any one of the standard answers, and determining the first correlation degree between the to-be-processed question and the any one of the standard answers according to the first statistical value; wherein the first statistical value comprises at least one of a maximum value, a minimum value, a mean value and a median value.

3. The method of claim 1, wherein, The reference answer is a plurality of, and there are a plurality of first correlation degrees between the to-be-processed question and each standard answer, and the first correlation degree is determined according to the text similarity between one reference answer and one standard answer. The determining the target answer from the plurality of standard answers according to the first correlation degrees of the plurality of standard answers comprises: for any one of the plurality of reference answers, determining an answer set from the plurality of standard answers based on the first correlation degrees determined according to the text similarities between the any one of the reference answers and the plurality of standard answers; wherein the answer set comprises at least one standard answer, and the first correlation degree determined according to the text similarity between the at least one standard answer and the any one of the reference answers is higher than a correlation degree threshold; determining the target answer from the answer sets corresponding to the plurality of reference answers according to the occurrence frequencies of the standard answers in the answer sets.

4. The method of claim 1, wherein, The performing answer prediction on the to-be-processed question by using a first language model to obtain at least one reference answer comprises: obtaining first prompt information, wherein the first prompt information is used to indicate task information to be processed by the first language model and a first number of reference answers output by the first language model; The first language model is used to predict answers of the to-be-processed question based on the first prompt information, so as to obtain the first quantity of reference answers.

5. The method of any one of claims 1-4, wherein, The target answer is multiple, and the reply to the to-be-processed question according to the target answer comprises: Obtaining a second correlation degree between the multiple target answers and the to-be-processed question; Based on the second correlation degree of the multiple target answers, the multiple target answers are sorted to obtain a sorting sequence; According to the target answer sorted in the front in the sorting sequence, the to-be-processed question is replied.

6. The method of claim 5, wherein, The second correlation degree between the multiple target answers and the to-be-processed question comprises: For any target answer, user feedback information listened to in at least one dimension of the any target answer is obtained, wherein each dimension corresponds to one user interaction behavior; According to the user feedback information in the at least one dimension, a score of the any target answer in the at least one dimension is determined; According to the score of the any target answer in the at least one dimension, the second correlation degree between the any target answer and the to-be-processed question is determined.

7. The method of claim 5, wherein, The second correlation degree between the multiple target answers and the to-be-processed question comprises: Obtaining third prompt information, wherein the third prompt information is used to prompt task information to be processed by a third language model; Using the third language model to predict the correlation degree of the to-be-processed question and the multiple target answers based on the third prompt information, so as to obtain a correlation degree score between the to-be-processed question and the multiple target answers; According to the correlation degree score of the multiple target answers, the second correlation degree between the multiple target answers and the to-be-processed question is determined.

8. The method of claim 5, wherein, The second correlation degree between the multiple target answers and the to-be-processed question comprises: Obtaining semantic similarity between the to-be-processed question and the multiple target answers; Based on the semantic similarity of the multiple target answers, the second correlation degree between the multiple target answers and the to-be-processed question is determined.

9. An artificial intelligence-based question processing device, comprising: A first obtaining module for obtaining a to-be-processed question; A first determining module for determining a first correlation degree between the to-be-processed question and multiple standard answers in an answer library; A second determining module for determining a target answer from the multiple standard answers according to the first correlation degree of the multiple standard answers; A reply module for replying to the to-be-processed question according to the target answer; The first determining module is configured to: Use a first language model to predict answers of the to-be-processed question to obtain at least one reference answer; Obtain a text similarity between the reference answer and the multiple standard answers; According to the text similarity between the reference answer and the multiple standard answers, the first correlation degree between the to-be-processed question and the multiple standard answers is determined.

10. The apparatus of claim 9, wherein, The reference answer is multiple, and there is a first correlation degree between the to-be-processed question and each standard answer; The first determining module is configured to: For any one of the multiple standard answers, the text similarity of the multiple reference answers and the any one standard answer is fused to obtain a first fused similarity, and a first correlation degree between the to-be-processed question and the any one standard answer is determined according to the first fused similarity. Or, For any one of the multiple standard answers, a first statistical value of the text similarity of the multiple reference answers and the any one standard answer is obtained, and a first correlation degree between the to-be-processed question and the any one standard answer is determined according to the first statistical value; wherein the first statistical value includes at least one of the maximum value, the minimum value, the mean value and the median value.

11. The apparatus of claim 9, wherein, The reference answers are multiple, and there are multiple first correlation degrees between the to-be-processed question and each standard answer, and the first correlation degree is determined according to the text similarity between one reference answer and one standard answer. The second determination module is configured to: For any one of the multiple reference answers, an answer set is determined from the multiple standard answers based on the first correlation degrees determined by the text similarity between the any one reference answer and the multiple standard answers; wherein the answer set includes at least one standard answer, and the first correlation degree determined by the text similarity between the at least one standard answer and the any one reference answer is higher than a correlation degree threshold value; According to the occurrence times of each standard answer in the answer set corresponding to the multiple reference answers, a target answer is determined from the answer set corresponding to the multiple reference answers.

12. The apparatus of claim 9, wherein, The first determination module is configured to: Obtain first prompt information, wherein the first prompt information is used to indicate task information to be processed by the first language model and a first number of reference answers output by the first language model; Using the first language model, the first number of reference answers are obtained by performing answer prediction on the to-be-processed question based on the first prompt information.

13. The apparatus of any of claims 9-12, wherein, The target answers are multiple, and the reply module is configured to: Obtain a second correlation degree between the multiple target answers and the to-be-processed question; Sort the multiple target answers based on the second correlation degrees of the multiple target answers to obtain a sorting sequence; Reply to the to-be-processed question according to the target answers sorted in the front in the sorting sequence.

14. The apparatus of claim 13, wherein, The reply module is configured to: For any one target answer, obtain user feedback information listened to by the any one target answer in at least one dimension, wherein each dimension corresponds to one user interaction behavior; Determine a score of the any one target answer in the at least one dimension according to the user feedback information in the at least one dimension; Determine a second correlation degree between the any one target answer and the to-be-processed question according to the score of the any one target answer in the at least one dimension.

15. The apparatus of claim 13, wherein, The reply module is configured to: Obtain third prompt information, wherein the third prompt information is used to prompt task information to be processed by a third language model; The third language model is used to predict the relevance of the to-be-processed question and the plurality of target answers based on the third prompt information, to obtain a relevance score between the to-be-processed question and the plurality of target answers. According to the relevance scores of the plurality of target answers, a second correlation degree between the plurality of target answers and the to-be-processed question is determined.

16. The apparatus of claim 13, wherein, The reply module is configured to: Obtain semantic similarity between the to-be-processed question and the plurality of target answers; Based on the semantic similarity of the plurality of target answers, a second correlation degree between the plurality of target answers and the to-be-processed question is determined.

17. An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the artificial intelligence-based question processing method of any one of claims 1-8.

18. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the artificial intelligence-based question processing method of any one of claims 1-8.

19. A computer program product comprising a computer program which, when executed by a processor, implements the steps of the artificial intelligence-based question processing method of any one of claims 1-8.

Images