A method and apparatus for handling a consultation question
By breaking down customer inquiries into multiple sub-questions through syntactic analysis and providing responses to each sub-question, the problem of inaccurate sentence vector matching in existing technologies is solved, resulting in more accurate inquiry responses.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INDUSTRIAL AND COMMERCIAL BANK OF CHINA
- Filing Date
- 2022-08-26
- Publication Date
- 2026-06-23
Smart Images
Figure CN115374263B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and specifically to a method and apparatus for processing consultation questions. Background Technology
[0002] With the development of natural language processing technology, intelligent question answering systems have emerged, and related technologies have been applied to many industries and fields.
[0003] In existing technologies, customer inquiries are segmented into words, and the word vectors of each segmented word are summed to obtain the sentence vector corresponding to the inquiry. Then, based on this sentence vector, the answer corresponding to the most similar basic question in a question-and-answer database is selected as the matching answer. However, when a customer's inquiry involves multiple basic questions, the existing method results in discrepancies between the sentence vector for the inquiry and the sentence vectors corresponding to the multiple basic questions during matching. This leads to an inability to accurately understand the user's intent, typically resulting in only a partial answer and an inaccurate result. Summary of the Invention
[0004] In view of the problems in the prior art, the present invention provides a method and apparatus for processing consultation problems, which can at least partially solve the problems existing in the prior art.
[0005] In a first aspect, the present invention proposes a method for processing consultation questions, comprising:
[0006] Receive a consultation request sent by a client, the consultation request including a consultation question;
[0007] Perform syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question;
[0008] If it is determined that there are verbs with a parallel relationship in the syntactic analysis results corresponding to the consultation question, then the consultation question is split into multiple sub-questions based on the verbs with a parallel relationship;
[0009] Based on each sub-question and the basic question-and-answer database, obtain the corresponding answer for each sub-question;
[0010] Return the answers to each sub-question to the client.
[0011] Secondly, the present invention provides an apparatus for processing consultation questions, comprising:
[0012] A receiving module is used to receive consultation requests sent by clients, the consultation requests including consultation questions;
[0013] The analysis module is used to perform syntactic analysis on the consultation question and obtain the syntactic analysis result corresponding to the consultation question;
[0014] The splitting module is used to split the consultation question into multiple sub-questions based on the verbs with a parallel relationship in the syntactic analysis results corresponding to the consultation question.
[0015] The module is used to obtain the answer to each sub-question based on each sub-question and the basic question-and-answer database.
[0016] The return module is used to return the answers to each sub-question to the client.
[0017] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for processing consultation problems described in any of the above embodiments.
[0018] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for processing consultation questions as described in any of the above embodiments.
[0019] Fifthly, the present invention provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the method for processing consultation problems as described in any of the above embodiments.
[0020] The method and apparatus for processing consultation questions provided in this invention receive a consultation request sent by a client, the consultation request including a consultation question; perform syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question; if it is determined that there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, then the consultation question is split into multiple sub-questions based on the verbs with a parallel relationship; obtain the answer corresponding to each sub-question according to each sub-question and a basic question-and-answer database; and return the answer corresponding to each sub-question to the client. This method can split the consultation question into multiple sub-questions and answer each sub-question, thereby improving the accuracy of consultation question responses. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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. In the drawings:
[0022] Figure 1 This is a flowchart illustrating the method for handling consultation issues provided in the first embodiment of the present invention.
[0023] Figure 2 This is a schematic diagram of the syntactic analysis results provided in the second embodiment of the present invention.
[0024] Figure 3 This is a flowchart illustrating the method for handling consultation issues provided in the third embodiment of the present invention.
[0025] Figure 4 This is a flowchart illustrating the method for handling consultation issues provided in the fourth embodiment of the present invention.
[0026] Figure 5 This is a flowchart illustrating the method for handling consultation issues provided in the fifth embodiment of the present invention.
[0027] Figure 6 This is a flowchart illustrating the method for handling consultation issues provided in the sixth embodiment of the present invention.
[0028] Figure 7 This is a flowchart illustrating the method for handling consultation issues provided in the seventh embodiment of the present invention.
[0029] Figure 8 This is a flowchart illustrating the method for handling consultation issues provided in the eighth embodiment of the present invention.
[0030] Figure 9 This is a schematic diagram of the structure of the consultation problem processing device provided in the ninth embodiment of the present invention.
[0031] Figure 10 This is a schematic diagram of the structure of the consultation problem processing device provided in the tenth embodiment of the present invention.
[0032] Figure 11 This is a schematic diagram of the structure of the consultation problem processing device provided in the eleventh embodiment of the present invention.
[0033] Figure 12 This is a schematic diagram of the structure of the consultation problem processing device provided in the twelfth embodiment of the present invention.
[0034] Figure 13 This is a schematic diagram of the structure of the consultation problem processing device provided in the thirteenth embodiment of the present invention.
[0035] Figure 14 This is a schematic diagram of the structure of the consultation problem processing device provided in the fourteenth embodiment of the present invention.
[0036] Figure 15 This is a schematic diagram of the physical structure of the electronic device provided in the fifteenth embodiment of the present invention. Detailed Implementation
[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.
[0038] To facilitate understanding of the technical solution provided in this application, the relevant content of the technical solution in this application will be explained below.
[0039] The consultation question processing method provided in this embodiment of the invention addresses situations where a customer's consultation question includes multiple questioning intentions. Based on syntactic analysis, it breaks down multiple questioning intentions into multiple sub-questions and performs matching of basic questions on each sub-question, thereby improving the accuracy of consultation question responses.
[0040] The following describes the specific implementation process of the consultation problem processing method provided in the embodiments of the present invention, taking the server as the execution subject as an example.
[0041] Figure 1 This is a flowchart illustrating the method for handling consultation issues provided in the first embodiment of the present invention, as shown below. Figure 1 As shown, the method for processing consultation questions provided in this embodiment of the invention includes:
[0042] S101. Receive a consultation request sent by the client, the consultation request including a consultation question;
[0043] Specifically, when a customer wants to consult about a problem, they can send a consultation request to the server through a client. The consultation request includes the question. The server will receive the consultation request. The client includes, but is not limited to, devices such as desktop computers, laptops, tablets, and smartphones.
[0044] For example, if customer X wants to know how the returns are calculated for a 3-month fixed-income open-ended wealth management product and whether it can be withdrawn early, customer X can log in to online banking using a laptop and enter "How are the returns calculated for a 3-month fixed-income open-ended wealth management product and whether it can be withdrawn early" as their inquiry on the intelligent customer service page. The laptop will then send an inquiry request carrying the above inquiry question to the online banking server, which will receive the inquiry request.
[0045] S102. Perform syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question;
[0046] Specifically, after obtaining the consultation question, the server can perform syntactic analysis on the consultation question to obtain the syntactic analysis results corresponding to the consultation question. The syntactic analysis results include the parts of speech of each word in the consultation question and the relationships between the words. The relationships between the words in the syntactic analysis results include, but are not limited to, attributive-head relation (ATT), verb-object relation (VOB), subject-verb relation (SBV), adverbial-head relation (ADV), verb-complement relation (CMP), and coordinate relation (COO).
[0047] For example, the inquiry question is: How is the return calculated for a 3-month fixed-income open-ended wealth management product, and can it be withdrawn early? The server performs syntactic analysis on the above inquiry question, and the obtained syntactic analysis results are as follows: Figure 2 As shown, Figure 2 The relationships between the various words and the part of speech of each word are shown. In the syntactic analysis results: the core HED of the consultation question is the predicate "calculate," and the verbs "calculate" and "withdraw" are in a parallel relationship. WP indicates punctuation. For details on the relationships between other words, see [link to syntactic analysis]. Figure 2 This will not be elaborated upon here.
[0048] S103. If it is determined that there are verbs with a parallel relationship in the syntactic analysis results corresponding to the consultation question, then the consultation question is split into multiple sub-questions based on the verbs with a parallel relationship.
[0049] Specifically, the server determines whether there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, that is, whether the syntactic analysis result includes words with a parallel relationship, and the part of speech of words with a parallel relationship is verb. If there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, it means that the consultation question involves multiple questioning intentions. Then the server will split the consultation question into multiple sub-questions based on the verbs with a parallel relationship, and each verb will correspond to one sub-question.
[0050] S104. Based on each sub-question and the basic question-and-answer database, obtain the answer corresponding to each sub-question;
[0051] Specifically, after the server breaks down the inquiry into multiple sub-questions, it matches a basic question for each sub-question from the basic question-and-answer database, and uses the answer to the basic question as the response to the matched sub-question, thus obtaining the response to each sub-question.
[0052] S105. Return the answers to each sub-question to the client.
[0053] Specifically, the server returns the answers to each sub-question as response information to the consultation request to the client terminal for the client to view.
[0054] The method for processing consultation questions provided in this invention includes receiving a consultation request sent by a client, the consultation request including a consultation question; performing syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question; if it is determined that there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, then splitting the consultation question into multiple sub-questions based on the verbs with a parallel relationship; obtaining the answer corresponding to each sub-question according to each sub-question and a basic question-and-answer database; and returning the answer corresponding to each sub-question to the client. This method can split the consultation question into multiple sub-questions and answer each sub-question, thereby improving the accuracy of consultation question responses.
[0055] Based on the above embodiments, further, the step of breaking down the consultation question into multiple sub-questions based on verbs with parallel relationships includes:
[0056] Each verb in the vocabulary of verbs with a parallel relationship and the modifiers corresponding to each verb are combined to form a sub-problem.
[0057] Specifically, the server retrieves each verb from a vocabulary of verbs with a parallel relationship, and then reassembles each verb and its corresponding modifiers into a sub-question, with each verb corresponding to one sub-question. The modifiers for each verb refer to nouns with a subject-predicate relationship and words that directly or indirectly modify the verb, obtained based on syntactic analysis results. All verbs in the vocabulary of verbs with a parallel relationship share the same subject; that is, if the nouns with a subject-predicate relationship and words that directly or indirectly modify the nouns for one verb in the vocabulary of verbs with a parallel relationship are obtained, then these nouns and words that directly or indirectly modify the nouns are used as modifiers for other verbs with a parallel relationship with that verb.
[0058] For example, Figure 2 In this context, the verbs with a parallel relationship are "calculate" and "withdraw". There is a subject-verb relationship between "product" and "calculate". The words that directly or indirectly modify "product" are: fixed-income, income-type, 3, each, month, fixed-term, open-ended, and wealth management. The words that directly or indirectly modify "calculate" are: how and income. Combining "calculate" and its corresponding modifiers into a sub-question: How is the income calculated for a 3-month fixed-income open-ended wealth management product?
[0059] "Withdrawal" and "calculation" are parallel terms. We can consider "product" and any words that directly or indirectly modify "product" as modifiers of "withdrawal." These modifiers include: whether, can, and early withdrawal. We can then group "withdrawal" and its corresponding modifiers into a sub-question: Can a 3-month fixed-income open-ended wealth management product be withdrawn early?
[0060] Figure 3 This is a flowchart illustrating the method for handling consultation issues provided in the third embodiment of the present invention, as shown below. Figure 3 As shown, based on the above embodiments, further, obtaining the answer corresponding to each sub-question based on each sub-question and the basic question-and-answer database includes:
[0061] S301. Segment the sub-problem into words to obtain the vocabulary corresponding to the sub-problem;
[0062] Specifically, the server performs word segmentation on the sub-problem to obtain the corresponding vocabulary. Word segmentation can be implemented using word segmentation tools, such as Jieba, PkuSeg, THULAC, etc., and the choice is made according to actual needs. This embodiment of the invention does not impose any limitations.
[0063] S302. Obtain the sentence vector corresponding to the sub-problem based on the vocabulary corresponding to the sub-problem;
[0064] Specifically, the server can obtain the sentence vector corresponding to the sub-question based on the vocabulary corresponding to the sub-question. This can be achieved by directly obtaining the sentence vector based on each word in the vocabulary corresponding to the sub-question; alternatively, by filtering out interfering words from the vocabulary corresponding to the sub-question and obtaining the sentence vector based on the remaining words, thus improving the accuracy of subsequent matching with the basic question.
[0065] For example, nouns and verbs are retained from the vocabulary corresponding to the sub-problem, and words other than nouns and verbs are filtered out. The sentence vector corresponding to the sub-problem is obtained through the nouns and verbs in the vocabulary corresponding to the sub-problem.
[0066] S303. Based on the sentence vectors corresponding to the sub-problems and the sentence vectors corresponding to each basic problem, obtain the basic problems that match the sub-problems; wherein, the sentence vectors corresponding to each basic problem are obtained in advance;
[0067] Specifically, the server compares the sentence vector corresponding to the sub-question with the sentence vector corresponding to each basic question, and obtains the basic question with the highest similarity to the sub-question as the basic question matching the sub-question. The sentence vector corresponding to each basic question is obtained in advance, and the basic questions are pre-set and stored in a basic question-answering database.
[0068] It is understandable that the sentence vectors for each basic problem are obtained in the same way as those for the sub-problems.
[0069] S304. Obtain the answer to the basic question that matches the sub-question from the basic question-answer database, and use it as the response to the sub-question.
[0070] Specifically, after obtaining the base question that matches the sub-question, the server queries the base question database for the answer corresponding to the base question that matches the sub-question, and uses this answer as the response to the sub-question. The base question-answer database is pre-defined and includes multiple question-answer pairs, each pair consisting of a base question and its corresponding answer.
[0071] Figure 4 This is a flowchart illustrating the method for handling consultation issues provided in the fourth embodiment of the present invention, as shown below. Figure 4 As shown, based on the above embodiments, further, obtaining the sentence vector corresponding to the sub-question based on the vocabulary corresponding to the sub-question includes:
[0072] S401. Based on the vocabulary corresponding to the sub-problem and the professional domain vocabulary similarity model, obtain the similarity score of each word in the vocabulary corresponding to the sub-problem; wherein, the professional domain vocabulary similarity model is obtained by training based on professional domain vocabulary classification training data and corresponding classification labels;
[0073] Specifically, the server uses a professional domain vocabulary similarity model to score each word in the vocabulary corresponding to the sub-question, determining whether it belongs to a specific professional domain. The higher the similarity score, the more likely the word is to belong to a professional domain. The professional domain vocabulary similarity model is trained based on professional domain vocabulary classification training data and corresponding classification labels. It is used to classify words and determine whether they belong to a specific professional domain or not. Classification labels are set according to actual needs, and this embodiment of the invention does not impose limitations. In this embodiment, a specific professional domain refers to a particular field, such as the financial field, and is set according to actual needs; this embodiment of the invention does not impose limitations.
[0074] For example, the category labels are 1 and 0, where 1 represents a specific professional field and 0 represents a non-specific professional field.
[0075] S402. Based on the similarity score and similarity threshold of each word in the vocabulary corresponding to the sub-question, obtain the retained vocabulary corresponding to the sub-question;
[0076] Specifically, the server compares the similarity score of each word in the vocabulary corresponding to the sub-question with a similarity threshold, removes words with similarity scores lower than the threshold, and retains words with similarity scores greater than or equal to the threshold, thus obtaining the retained vocabulary corresponding to the sub-question. The retained vocabulary consists of important words that can express the questioning intent of the sub-question.
[0077] S403. Based on the reserved words corresponding to the sub-question and the similarity scores of each reserved word, obtain the sentence vector corresponding to the sub-question.
[0078] Specifically, the server can obtain the word vector corresponding to each retained word in the retained vocabulary corresponding to the sub-question, use the similarity score of each retained word as the weight of the word vector corresponding to each retained word, and sum the word vectors corresponding to each retained word in a weighted manner to obtain the sentence vector corresponding to the sub-question. The conversion of words into vectors can be achieved using the Word2vec model, a word vector model used to map words to vectors.
[0079] For example, the sub-problem has 10 reserved words, and the word vector corresponding to each reserved word is denoted as w. i The similarity score for each retained word is denoted as p. i The sentence vector corresponding to the sub-problem is: Where i is a positive integer and i is less than or equal to 10.
[0080] Figure 5 This is a flowchart illustrating the method for handling consultation issues provided in the fifth embodiment of the present invention, as shown below. Figure 5 As shown, based on the above embodiments, the further step of training the professional domain vocabulary similarity model based on professional domain vocabulary classification training data and corresponding classification labels includes:
[0081] S501. Obtain training data for professional domain vocabulary classification and corresponding classification labels;
[0082] Specifically, vocabulary related to specific professional fields is collected, and corresponding category labels are set as belonging to those specific professional fields; vocabulary related to non-specific professional fields is also collected, and corresponding category labels are set as belonging to those non-specific professional fields. The collected vocabulary related to specific and non-specific professional fields is used as training data for professional field vocabulary classification. The server can obtain the professional field vocabulary classification training data and its corresponding category labels. The number of training words included in the professional field vocabulary classification training data is selected according to actual needs, and this embodiment of the invention does not impose a limitation.
[0083] For example, for the financial field, financial vocabulary is collected and categorized as "financial"; non-financial vocabulary is collected and categorized as "non-financial". The collected financial and non-financial vocabulary is used as training data for financial vocabulary classification. To improve the efficiency of financial vocabulary collection, questions in a financial question-and-answer database are segmented to obtain a first vocabulary set; questions in a publicly available casual chat question-and-answer database are segmented to obtain a second vocabulary set, and all words in the second set are considered non-financial vocabulary. The intersection of the first and second vocabulary sets is taken to obtain a third vocabulary set. The third vocabulary set is removed from the first vocabulary set, and the remaining words are considered financial vocabulary. The financial and casual chat question-and-answer databases are existing databases. The obtained financial and non-financial vocabulary can be corrected by experts.
[0084] S502. Convert each training word in the professional domain vocabulary classification training data into a vector to obtain the sample features corresponding to each training word.
[0085] Specifically, the server converts each training word in the professional domain vocabulary classification training data into a vector, thereby obtaining the sample features corresponding to each training word. The conversion method used to transform the training words into vectors can be BERT, word2vec, etc., and the choice is made according to actual needs; this embodiment of the invention does not impose any limitations.
[0086] S503. Based on the original model, the sample features corresponding to each training word, and their respective classification labels, a professional domain word similarity model is trained.
[0087] Specifically, the server trains the original model based on the sample features and classification labels corresponding to each training word, thereby obtaining a domain-specific vocabulary similarity model. The original model can be selected according to actual needs, such as the FastText model, Random Forest model, XGBoost model, or TextCNN model. K-fold cross-validation can be used for both training and validation of the model.
[0088] Understandably, for a given word, a domain-specific vocabulary similarity model can output a probability value indicating that the word belongs to a specific domain, and this probability value serves as the word's similarity score. By setting a classification threshold, words with a probability value greater than or equal to the threshold are classified as domain-specific vocabulary, while words with a probability value less than the threshold are classified as non-domain-specific vocabulary. The classification threshold can be set according to actual needs, and this embodiment of the invention does not impose any limitations.
[0089] Figure 6 This is a flowchart illustrating the method for handling consultation issues provided in the sixth embodiment of the present invention, as shown below. Figure 6 As shown, based on the above embodiments, further, obtaining the answer corresponding to each sub-question based on each sub-question and the basic question-and-answer database includes:
[0090] S601. Segment the sub-problem into words to obtain the vocabulary corresponding to the sub-problem;
[0091] Specifically, the server performs word segmentation on the sub-problem to obtain the corresponding vocabulary. Word segmentation can be implemented using word segmentation tools, such as Jieba, PkuSeg, THULAC, etc., and the choice is made according to actual needs. This embodiment of the invention does not impose any limitations.
[0092] S602. Based on the vocabulary corresponding to the sub-problem and each similarity sub-model included in the professional domain vocabulary similarity model, obtain the similarity score of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model; wherein, the professional domain vocabulary similarity model includes multiple similarity sub-models, and each similarity sub-model is obtained by training based on professional domain vocabulary classification training data and corresponding classification labels;
[0093] Specifically, the server scores each word in the vocabulary corresponding to the sub-question based on whether it belongs to a specific professional domain through each similarity sub-model included in the professional domain vocabulary similarity model. The similarity score for each word in the vocabulary corresponding to the sub-question under each similarity sub-model can differ for the same word obtained through different similarity sub-models. The professional domain vocabulary similarity model includes multiple similarity sub-models, each trained based on professional domain vocabulary classification training data and corresponding classification labels.
[0094] S603. Based on the similarity score and similarity threshold of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model, obtain the retained vocabulary of the sub-problem under each similarity sub-model;
[0095] Specifically, the server compares the similarity score of each word in the vocabulary corresponding to the sub-problem with a similarity threshold under each similarity sub-model, removes words with similarity scores less than the similarity threshold, and retains words with similarity scores greater than or equal to the similarity threshold, thereby obtaining the retained vocabulary of the sub-problem under each similarity sub-model.
[0096] S604. Based on the retained words of the sub-problem under each similarity sub-model and the similarity score of each retained word under each similarity sub-model, obtain the sentence vector corresponding to the sub-problem;
[0097] Specifically, for each similarity sub-model, the server can obtain the word vector corresponding to each retained word in the retained vocabulary of the sub-question under the similarity sub-model. The similarity score of each retained word in the retained vocabulary of the sub-question under the similarity sub-model is used as the weight of the word vector corresponding to each retained word. The word vectors corresponding to each retained word are then weighted and summed to obtain a sentence vector corresponding to the sub-question. Multiple sentence vectors corresponding to the sub-question can be obtained, and the number of sentence vectors corresponding to the sub-question is the same as the number of similarity sub-models.
[0098] S605. Based on the sentence vector corresponding to each sub-problem and the sentence vector corresponding to each basic problem, obtain the highest similarity value between the sentence vector corresponding to each sub-problem and the sentence vector corresponding to each basic problem; wherein, the sentence vector corresponding to each basic problem is obtained in advance;
[0099] Specifically, the server calculates the similarity value between a sentence vector corresponding to the sub-question and the sentence vector corresponding to each basic question. The server then selects the highest similarity value from the similarity values between this sentence vector and the sentence vectors corresponding to each basic question, and uses this as the highest similarity value between the sentence vector and the sentence vector corresponding to each basic question. This process is repeated until each sentence vector among the multiple sentence vectors corresponding to the sub-question obtains a highest similarity value. The sentence vectors corresponding to each basic question are pre-obtained, and the basic questions are pre-set and stored in a basic question-answering database.
[0100] S606. Obtain the base problem corresponding to the highest highest similarity value among the highest similarity values between the sentence vectors corresponding to each sub-problem and the sentence vectors corresponding to the base problem, and use it as the base problem to match the sub-problem;
[0101] Specifically, the server compares the highest similarity value of each sentence vector corresponding to the sub-question with the highest similarity value of the sentence vector corresponding to the base question, obtains the highest highest similarity value, and uses the base question corresponding to the highest highest similarity value as the base question for matching the sub-question.
[0102] S607. Obtain the answer to the basic question that matches the sub-question from the basic question-answer database, and use it as the response to the sub-question.
[0103] Specifically, after obtaining the base question that matches the sub-question, the server queries the base question database for the answer corresponding to the base question that matches the sub-question, and uses this answer as the response to the sub-question. The base question-answer database is pre-defined and includes multiple question-answer pairs, each pair consisting of a base question and its corresponding answer.
[0104] Figure 7 This is a flowchart illustrating the method for handling consultation issues provided in the seventh embodiment of the present invention, as shown below. Figure 7 As shown, based on the above embodiments, the further step of training the professional domain vocabulary similarity model based on professional domain vocabulary classification training data and corresponding classification labels includes:
[0105] S701. Obtain training data for professional domain vocabulary classification and corresponding classification labels;
[0106] Specifically, the implementation process of this step is similar to that of step S501, and will not be described in detail here.
[0107] S702. Convert each training word in the professional domain vocabulary classification training data into a vector to obtain the sample features corresponding to each training word.
[0108] Specifically, the implementation process of this step is similar to that of step S502, and will not be described in detail here.
[0109] S703. Based on the M original models, the sample features corresponding to each training word, and their respective classification labels, train M similarity sub-models respectively; where M is a positive integer greater than or equal to 3.
[0110] Specifically, the server trains each original model based on the sample features and classification labels corresponding to each training word, thereby obtaining each similarity sub-model. Each original model corresponds to one similarity sub-model. Here, M is a positive integer greater than or equal to 3. The M original models can be FastText, Random Forest, XGBoost, TextCNN, etc., selected according to actual needs; this embodiment of the invention does not impose any limitations. K-fold cross-validation can be used for model training and validation.
[0111] S704. Select multiple similarity sub-models with an accuracy greater than the accuracy threshold from the M similarity sub-models to form the professional domain vocabulary similarity model.
[0112] Specifically, the server obtains the accuracy of each similarity sub-model, compares the accuracy of each of the M similarity sub-models with an accuracy threshold, and selects multiple similarity models with an accuracy greater than the accuracy threshold to form the domain-specific vocabulary similarity model. The accuracy of each similarity sub-model can be obtained during model validation to evaluate the accuracy of the similarity sub-model in vocabulary classification. The accuracy threshold is set according to actual needs, and this embodiment of the invention does not impose any limitations.
[0113] Understandably, if there is only one similarity sub-model with an accuracy greater than the accuracy threshold, then that similarity sub-model can be directly used as the domain-specific vocabulary similarity model. If there is no similarity sub-model with an accuracy greater than the accuracy threshold, then the parameters need to be adjusted and retrained.
[0114] Figure 8 This is a flowchart illustrating the method for handling consultation issues provided in the eighth embodiment of the present invention, as shown below. Figure 8 As shown, based on the above embodiments, the further step of obtaining the sentence vectors corresponding to each basic question in advance includes:
[0115] S801. Segment the basic questions to obtain the corresponding vocabulary.
[0116] Specifically, the implementation process of this step is similar to that of step S601, and will not be described in detail here.
[0117] S802. Based on the vocabulary corresponding to the basic question and each similarity sub-model included in the professional domain vocabulary similarity model, obtain the similarity score of each word in the vocabulary corresponding to the basic question under each similarity sub-model.
[0118] Specifically, the implementation process of this step is similar to that of step S602, and will not be described in detail here.
[0119] S803. Based on the similarity score and similarity threshold of each word in the vocabulary corresponding to the basic question under each similarity sub-model, obtain the retained vocabulary of the basic question under each similarity sub-model;
[0120] Specifically, the implementation process of this step is similar to that of step S603, and will not be described in detail here.
[0121] S804. Based on the vocabulary retained under each similarity sub-model of the basic question and the similarity score of each retained vocabulary under each similarity sub-model, obtain multiple sentence vectors corresponding to the basic question.
[0122] Specifically, the implementation process of this step is similar to that of step S604, and will not be described in detail here.
[0123] Understandably, when calculating the similarity between the sentence vectors corresponding to the sub-problems and the sentence vectors corresponding to the basic problems, the similarity sub-model used to obtain the sentence vectors corresponding to the sub-problems is the same as the similarity sub-model used to obtain the sentence vectors corresponding to the basic problems.
[0124] The following example illustrates the specific implementation process of the financial problem-solving method provided by the embodiments of the present invention.
[0125] A customer wants to know how the returns of a 3-month fixed-income open-ended wealth management product are calculated and whether it can be withdrawn early. They log into online banking on a laptop and enter "How are the returns calculated for a 3-month fixed-income open-ended wealth management product? Can it be withdrawn early?" as their question on the intelligent customer service's inquiry page.
[0126] The laptop will send an inquiry request carrying the question Q to the online banking server, and the online banking server will receive the inquiry request.
[0127] The server performs syntactic analysis on the consultation question Q and obtains the syntactic analysis results corresponding to consultation question Q, such as... Figure 2 As shown. Because the syntactic analysis result corresponding to consultation question Q contains the parallel relationships "calculate" and "withdraw", the server splits consultation question Q into the following two sub-questions based on "calculate" and "withdraw":
[0128] Sub-question 1: How to calculate the return on a 3-month fixed-income open-ended wealth management product;
[0129] Sub-question 2: Can fixed-income 3-month fixed-term open-ended wealth management products be withdrawn early?
[0130] The server performs word segmentation on sub-problem 1, obtaining the following vocabulary corresponding to sub-problem 1:
[0131] [Fixed], [Income-Rating], [3 Months], [Fixed-Term], [Open-Ended], [Wealth Management Products], [How], [Calculate], [Returns]
[0132] Based on the vocabulary corresponding to sub-question 1 and the vocabulary similarity model in the financial field, the server can obtain the similarity scores for each of the following: [fixed], [income-related], [3 months], [fixed term], [open type], [financial product], [how], [calculate], and [return].
[0133] Based on the similarity scores and similarity thresholds of the aforementioned words, the server obtains the following reserved words as sub-problem 1: [fixed], [income-related], [3 months], [fixed term], [open type], [financial product], [calculation], and [income]. Based on the reserved words of sub-problem 1 and the similarity scores of each reserved word, the server obtains the sentence vector corresponding to sub-problem 1.
[0134] Based on the sentence vector corresponding to sub-question 1 and the sentence vectors corresponding to each basic question, obtain the basic question that matches sub-question 1; the sentence vectors corresponding to each basic question are obtained in advance; the answer corresponding to the basic question that matches sub-question 1 is obtained from the basic question-answering database, and is used as the response to sub-question 1.
[0135] The specific process of obtaining the answer to sub-problem 2 is similar to that of obtaining the answer to sub-problem 1, and will not be elaborated here.
[0136] The server returns the answers to sub-question 1 and sub-question 2 to the laptop for the client to view.
[0137] The consultation question processing method provided in this invention addresses situations where a consultation question includes multiple questioning intentions. It breaks down these intentions into multiple sub-questions, each sub-question, and matches it with a basic question to obtain a response, thus improving the accuracy of consultation question answers. Furthermore, it filters out non-key words from the vocabulary corresponding to the sub-questions, retaining key words to obtain vectors for the sub-questions, improving the accuracy of question matching and consequently enhancing the accuracy of obtaining responses to the sub-questions.
[0138] Figure 9 This is a schematic diagram of the structure of the consultation problem processing device provided in the ninth embodiment of the present invention, as shown below. Figure 9 As shown, the consultation problem processing device provided in this embodiment of the invention includes a receiving module 910, an analysis module 920, a splitting module 930, an obtaining module 940, and a return module 950, wherein:
[0139] The receiving module 910 is used to receive a consultation request sent by a client, the consultation request including a consultation question; the analysis module 920 is used to perform syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question; the splitting module 930 is used to split the consultation question into multiple sub-questions based on the verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question after determining that there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question; the obtaining module 940 is used to obtain the answer corresponding to each sub-question based on each sub-question and the basic question-and-answer database; and the returning module 950 is used to return the answer corresponding to each sub-question to the client.
[0140] Specifically, when a customer wants to consult about a problem, they can send a consultation request, including the question, to the receiving module 910 through the client. The receiving module 910 will receive the consultation request. The client includes, but is not limited to, devices such as desktop computers, laptops, tablets, and smartphones.
[0141] After obtaining the consultation question, the analysis module 920 can perform syntactic analysis on the consultation question to obtain the syntactic analysis results corresponding to the consultation question. The syntactic analysis results include the parts of speech of each word in the consultation question and the relationships between each word. The relationships between each word in the syntactic analysis results include, but are not limited to, attributive-head relationships, verb-object relationships, subject-predicate relationships, adverbial-head relationships, verb-complement relationships, and coordinate relationships.
[0142] The splitting module 930 determines whether there are verb words with a parallel relationship in the syntactic analysis result corresponding to the consultation question, that is, whether the syntactic analysis result includes words with a parallel relationship, and whether the part of speech of words with a parallel relationship is verb. If there are verb words with a parallel relationship in the syntactic analysis result corresponding to the consultation question, it means that the consultation question involves multiple questioning intentions. Then the splitting module 930 will split the consultation question into multiple sub-questions based on the verb words with a parallel relationship, and each verb word will correspond to one sub-question.
[0143] After breaking down the consultation question into multiple sub-questions, module 940 matches a basic question for each sub-question from the basic question-answer database, and uses the answer to the basic question as the reply to the matched sub-question, thus obtaining the reply to each sub-question.
[0144] The return module 950 returns the answers to each sub-question as response information to the consultation request to the client terminal for the client to view.
[0145] The consultation question processing apparatus provided in this embodiment of the invention receives a consultation request sent by a client, the consultation request including a consultation question; performs syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question; if it is determined that there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, then the consultation question is split into multiple sub-questions based on the verbs with a parallel relationship; according to each sub-question and a basic question-and-answer database, the answer corresponding to each sub-question is obtained; and the answer corresponding to each sub-question is returned to the client. This apparatus can split the consultation question into multiple sub-questions and answer each sub-question, thereby improving the accuracy of consultation question responses.
[0146] Based on the above embodiments, further, the step of breaking down the consultation question into multiple sub-questions based on verbs with parallel relationships includes:
[0147] Each verb in the vocabulary of verbs with a parallel relationship and the modifiers corresponding to each verb are combined to form a sub-problem.
[0148] Figure 10 This is a schematic diagram of the structure of the consultation problem processing device provided in the tenth embodiment of the present invention, as shown below. Figure 10 As shown, based on the above embodiments, the obtaining module 940 further includes a first word segmentation unit 9401, a first obtaining unit 9402, a second obtaining unit 9403, and a first acquisition unit 9404, wherein:
[0149] The first word segmentation unit 9401 is used to segment the sub-question into words to obtain the vocabulary corresponding to the sub-question; the first obtaining unit 9402 is used to obtain the sentence vector corresponding to the sub-question based on the vocabulary corresponding to the sub-question; the second obtaining unit 9403 is used to obtain the basic question matching the sub-question based on the sentence vector corresponding to the sub-question and the sentence vectors corresponding to each basic question; wherein, the sentence vectors corresponding to each basic question are obtained in advance; the first obtaining unit 9404 is used to obtain the answer corresponding to the basic question matching the sub-question from the basic question-answering database as the response to the sub-question.
[0150] Figure 11 This is a schematic diagram of the structure of the consultation problem processing method provided in the eleventh embodiment of the present invention, as shown below. Figure 11 As shown, based on the above embodiments, the first obtaining unit 9402 further includes a first obtaining subunit 94021, a second obtaining subunit 94022, and a third obtaining subunit 94023, wherein:
[0151] The first obtaining subunit 94021 is used to obtain the similarity score of each word in the vocabulary corresponding to the sub-problem based on the vocabulary corresponding to the sub-problem and the professional domain vocabulary similarity model; wherein, the professional domain vocabulary similarity model is obtained by training based on professional domain vocabulary classification training data and corresponding classification labels; the second obtaining subunit 94022 is used to obtain the retained vocabulary corresponding to the sub-problem based on the similarity score of each word in the vocabulary corresponding to the sub-problem and the similarity threshold; the third obtaining subunit 94023 is used to obtain the sentence vector corresponding to the sub-problem based on the retained vocabulary corresponding to the sub-problem and the similarity score of each retained word.
[0152] Figure 12 This is a schematic diagram of the structure of the consultation problem processing method provided in the twelfth embodiment of the present invention, as shown below. Figure 12 As shown, based on the above embodiments, the first obtaining unit 9402 further includes an obtaining subunit 94024, a conversion subunit 94025, and a training subunit 94026:
[0153] The acquisition subunit 94024 is used to acquire professional domain vocabulary classification training data and corresponding classification labels; the transformation subunit 94025 is used to convert each training word in the professional domain vocabulary classification training data into a vector to obtain the sample features corresponding to each training word; the training subunit 94026 is used to train a professional domain vocabulary similarity model based on the original model, the sample features corresponding to each training word and their respective classification labels.
[0154] Figure 13 This is a schematic diagram of the structure of the consultation problem processing method provided in the thirteenth embodiment of the present invention, as shown below. Figure 13 As shown, based on the above embodiments, the obtaining module 940 further includes a second word segmentation unit 9405, a third obtaining unit 9406, a fourth obtaining unit 9407, a fifth obtaining unit 9408, a sixth obtaining unit 9409, a second obtaining unit 9410, and a third obtaining unit 9411, wherein:
[0155] The second word segmentation unit 9405 is used to segment the sub-problem to obtain the vocabulary corresponding to the sub-problem; the third obtaining unit 9406 is used to obtain the similarity score of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model based on the vocabulary corresponding to the sub-problem and each similarity sub-model included in the professional domain vocabulary similarity model; wherein, the professional domain vocabulary similarity model includes multiple similarity sub-models, each similarity sub-model is obtained by training based on professional domain vocabulary classification training data and corresponding classification labels; the fourth obtaining unit 9407 is used to obtain the retained vocabulary of the sub-problem under each similarity sub-model based on the similarity score of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model and the similarity threshold; the fifth obtaining unit 9408 is used to obtain the retained vocabulary of the sub-problem under each similarity sub-model based on the similarity score of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model. The system obtains the retained vocabulary under the model and the similarity score of each retained vocabulary under each similarity sub-model to obtain the sentence vector corresponding to the sub-question; the sixth obtaining unit 9409 is used to obtain the highest similarity value between each sentence vector corresponding to the sub-question and the sentence vector corresponding to each basic question based on each sentence vector corresponding to the sub-question and the sentence vector corresponding to each basic question; wherein, the sentence vector corresponding to each basic question is obtained in advance; the second obtaining unit 9410 is used to obtain the basic question corresponding to the highest similarity value among the highest similarity values between each sentence vector corresponding to the sub-question and the sentence vector corresponding to the basic question, as the basic question matching the sub-question; the third obtaining unit 9411 is used to obtain the answer corresponding to the basic question matching the sub-question from the basic question-answering library, as the answer corresponding to the sub-question.
[0156] Figure 14 This is a schematic diagram of the structure of the consultation problem processing method provided in the fourteenth embodiment of the present invention, as shown below. Figure 14 As shown, based on the above embodiments, the obtaining module 940 further includes a fourth obtaining unit 9412, a conversion unit 9413, a training unit 9414, and a composition unit 9415, wherein:
[0157] The fourth acquisition unit 9412 is used to acquire professional domain vocabulary classification training data and corresponding classification labels; the conversion unit 9413 is used to convert each training word in the professional domain vocabulary classification training data into a vector to obtain the sample features corresponding to each training word; the training unit 9414 is used to train and obtain M similarity sub-models according to M original models, the sample features corresponding to each training word and their respective classification labels; where M is a positive integer greater than or equal to 3; the composition unit 9415 is used to select multiple similarity sub-models with an accuracy greater than the accuracy threshold from the M similarity sub-models to form the professional domain vocabulary similarity model.
[0158] The embodiments of the device provided in this invention can be used to execute the processing flow of the above-described method embodiments. Its functions will not be repeated here, but can be referred to the detailed description of the above-described method embodiments.
[0159] It should be noted that the method and apparatus for handling consultation issues provided in the embodiments of the present invention can be used in the financial field, or in any technical field other than the financial field. The embodiments of the present invention do not limit the application field of the method and apparatus for handling consultation issues.
[0160] Figure 15 This is a schematic diagram of the physical structure of the electronic device provided in the fifteenth embodiment of the present invention, as shown below. Figure 15 As shown, the electronic device may include a processor 1501, a communications interface 1502, a memory 1503, and a communication bus 1504. The processor 1501, communications interface 1502, and memory 1503 communicate with each other via the communication bus 1504. The processor 1501 can call logical instructions in the memory 1503 to execute the following methods: receiving a consultation request sent by a client, the consultation request including a consultation question; performing syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question; if it is determined that there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, then splitting the consultation question into multiple sub-questions based on the verbs with a parallel relationship; obtaining the answer corresponding to each sub-question based on each sub-question and a basic question-and-answer database; and returning the answers corresponding to each sub-question to the client.
[0161] Furthermore, the logical instructions in the aforementioned memory 1503 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0162] This embodiment discloses a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer can execute the methods provided in the above-described method embodiments, such as: performing syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question; if it is determined that there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, then splitting the consultation question into multiple sub-questions based on the verbs with a parallel relationship; obtaining the answer corresponding to each sub-question according to each sub-question and a basic question-and-answer database; and returning the answer corresponding to each sub-question to the client.
[0163] This embodiment provides a computer-readable storage medium storing a computer program that causes a computer to execute the methods provided in the above-described method embodiments. For example, the methods include: performing syntactic analysis on the consultation question to obtain a syntactic analysis result corresponding to the consultation question; if it is determined that there are verbs with a parallel relationship in the syntactic analysis result corresponding to the consultation question, then splitting the consultation question into multiple sub-questions based on the verbs with a parallel relationship; obtaining a response corresponding to each sub-question based on each sub-question and a basic question-and-answer database; and returning the responses corresponding to each sub-question to the client.
[0164] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0165] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0166] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0167] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0168] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0169] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for handling consultation issues, characterized in that, include: Receive a consultation request sent by a client, the consultation request including a consultation question; Perform syntactic analysis on the consultation question to obtain the syntactic analysis result corresponding to the consultation question; If it is determined that there are verbs with a parallel relationship in the syntactic analysis results corresponding to the consultation question, then the consultation question is split into multiple sub-questions based on the verbs with a parallel relationship; Based on each sub-question and the basic question-and-answer database, obtain the corresponding answer for each sub-question; Return the answers to each sub-question to the client; The step of obtaining the answer to each sub-question based on each sub-question and the basic question-and-answer database includes: Based on the vocabulary corresponding to the sub-question and each similarity sub-model included in the professional domain vocabulary similarity model, the similarity score of each word in the vocabulary corresponding to the sub-question under each similarity sub-model is obtained; wherein, the professional domain vocabulary similarity model includes multiple similarity sub-models, and each similarity sub-model is obtained by training based on professional domain vocabulary classification training data and corresponding classification labels; Based on the similarity score and similarity threshold of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model, the retained vocabulary of the sub-problem under each similarity sub-model is obtained; Based on the vocabulary retained under each similarity sub-model for the sub-question and the similarity score of each retained vocabulary under each similarity sub-model, obtain the sentence vector corresponding to the sub-question; Based on the sentence vector corresponding to each sub-problem and the sentence vector corresponding to each basic problem, the highest similarity value between the sentence vector corresponding to each sub-problem and the sentence vector corresponding to each basic problem is obtained; wherein, the sentence vector corresponding to each basic problem is obtained in advance; Obtain the base question corresponding to the highest highest similarity value among the highest similarity values between the sentence vectors corresponding to each sub-problem and the sentence vectors corresponding to the base question, and use this base question as the matching base question for the sub-problem; The answer to the basic question that matches the sub-question is obtained from the basic question-answering database and used as the response to the sub-question. Specifically, obtaining each sentence vector corresponding to the sub-question based on the retained words in each similarity sub-model and the similarity score of each retained word in each similarity sub-model involves obtaining the word vector corresponding to each retained word in the retained words of the sub-question under the similarity sub-model, using the similarity score of each retained word in the retained words of the sub-question under the similarity sub-model as the weight of the word vector corresponding to each retained word, and weighting and summing the word vectors corresponding to each retained word to obtain a sentence vector corresponding to the sub-question.
2. The method according to claim 1, characterized in that, The method of breaking down the consultation question into multiple sub-questions based on verbs with parallel relationships includes: Each verb in the vocabulary of verbs with a parallel relationship and the modifiers corresponding to each verb are combined to form a sub-problem.
3. The method according to claim 1 or 2, characterized in that, The process of obtaining the answer for each sub-question based on each sub-question and the basic question-and-answer database includes: The sub-problem is segmented to obtain the vocabulary corresponding to the sub-problem; Based on the vocabulary corresponding to the sub-problem, obtain the sentence vector corresponding to the sub-problem; Based on the sentence vectors corresponding to the sub-problems and the sentence vectors corresponding to each basic problem, the basic problems that match the sub-problems are obtained; wherein, the sentence vectors corresponding to each basic problem are obtained in advance; The answer to the basic question that matches the sub-question is obtained from the basic question-answering database and used as the response to the sub-question.
4. The method according to claim 3, characterized in that, The step of obtaining the sentence vector corresponding to the sub-question based on the vocabulary corresponding to the sub-question includes: Based on the vocabulary corresponding to the sub-question and the professional domain vocabulary similarity model, a similarity score is obtained for each word in the vocabulary corresponding to the sub-question; wherein, the professional domain vocabulary similarity model is obtained by training based on professional domain vocabulary classification training data and corresponding classification labels; Based on the similarity score and similarity threshold of each word in the vocabulary corresponding to the sub-question, the retained vocabulary corresponding to the sub-question is obtained; Based on the reserved words corresponding to the sub-question and the similarity scores of each reserved word, the sentence vector corresponding to the sub-question is obtained.
5. The method according to claim 4, characterized in that, The steps for obtaining the professional domain vocabulary similarity model based on professional domain vocabulary classification training data and corresponding classification labels include: Acquire training data for classifying professional domain vocabulary and corresponding classification labels; Each training word in the professional domain vocabulary classification training data is converted into a vector to obtain the sample features corresponding to each training word. Based on the original model, the sample features corresponding to each training word, and their respective classification labels, a professional domain word similarity model is trained.
6. The method according to claim 1, characterized in that, The steps for obtaining the professional domain vocabulary similarity model based on professional domain vocabulary classification training data and corresponding classification labels include: Based on M original models, the sample features corresponding to each training word, and their respective classification labels, M similarity sub-models are trained respectively; where M is a positive integer greater than or equal to 3. Multiple similarity sub-models with an accuracy greater than the accuracy threshold are selected from M similarity sub-models to form the professional domain vocabulary similarity model.
7. A device for processing consultation questions, characterized in that, include: A receiving module is used to receive consultation requests sent by clients, the consultation requests including consultation questions; The analysis module is used to perform syntactic analysis on the consultation question and obtain the syntactic analysis result corresponding to the consultation question; The splitting module is used to split the consultation question into multiple sub-questions based on the verbs with a parallel relationship in the syntactic analysis results corresponding to the consultation question. The module is used to obtain the answer to each sub-question based on each sub-question and the basic question-and-answer database. The return module is used to return the answers to each sub-question to the client; The obtaining module includes a second word segmentation unit, a third obtaining unit, a fourth obtaining unit, a fifth obtaining unit, a sixth obtaining unit, a second acquisition unit, and a third acquisition unit, wherein: The second word segmentation unit is used to segment the sub-problem into words to obtain the vocabulary corresponding to the sub-problem; The third obtaining unit is used to obtain the similarity score of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model based on the vocabulary corresponding to the sub-problem and each similarity sub-model included in the professional domain vocabulary similarity model; wherein, the professional domain vocabulary similarity model includes multiple similarity sub-models, and each similarity sub-model is obtained by training based on professional domain vocabulary classification training data and corresponding classification labels; The fourth obtaining unit is used to obtain the retained words of the sub-problem under each similarity sub-model based on the similarity score and similarity threshold of each word in the vocabulary corresponding to the sub-problem under each similarity sub-model; The fifth obtaining unit is used to obtain the sentence vector corresponding to the sub-question based on the retained words of the sub-question under each similarity sub-model and the similarity score of each retained word under each similarity sub-model; The sixth obtaining unit is used to obtain the highest similarity value between each sentence vector corresponding to the sub-problem and the sentence vector corresponding to each basic problem, based on each sentence vector corresponding to the sub-problem and the sentence vector corresponding to each basic problem; wherein, the sentence vector corresponding to each basic problem is obtained in advance; The second acquisition unit is used to acquire the base question corresponding to the highest highest similarity value among the highest similarity values between the sentence vectors corresponding to each sub-problem and the sentence vectors corresponding to the base question, and use it as the base question to match the sub-problem; The third acquisition unit is used to acquire the answer to the basic question that matches the sub-question from the basic question-answer database, and use it as the response to the sub-question; Specifically, the fifth obtaining unit is used to obtain the word vector corresponding to each retained word in the retained vocabulary of the sub-problem under the similarity sub-model, take the similarity score of each retained word in the retained vocabulary of the sub-problem under the similarity sub-model as the weight of the word vector corresponding to each retained word, and add the word vectors corresponding to each retained word in a weighted sum to obtain a sentence vector corresponding to the sub-problem.
8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6.
10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6.