Session information processing method, apparatus and system, and statement generation method and apparatus

Abstract

The present application provides a session information processing method, apparatus and system, and a statement generation method and apparatus. The session information processing method comprises: receiving a first input of a user, wherein the first input comprises first session information; in response to the first input, carrying out matching on the first session information in a first database; when the matching is successful, inputting, into a first model, second session information and first reply information that correspond to the first session information, so that the first model outputs second reply information on the basis of the second session information and the first reply information; and displaying the second reply information, wherein the first database is a professional knowledge base in the professional field to which the first session information belongs, and the first model is an intelligent dialogue model.

Description

Method, device and system for processing conversation information, and method and device for generating sentences

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on June 30, 2023, with application number "202310795163.2" and application name "Sentence Generation Method, Device, Cooking Equipment and Readable Storage Medium", the entire contents of which are incorporated by reference into this application.

[0002] This application claims priority to the Chinese patent application filed with the China Patent Office on July 19, 2023, with application number "202310891118.7" and application name "Method, device and system for processing session information, electronic device and storage medium", the entire contents of which are incorporated by reference into this application. Technical Field

[0003] The present application relates to the field of intelligent question-answering technology, and more specifically, to a method, device, and system for processing conversation information, and a method and device for generating sentences. Background Art

[0004] Currently, for question-and-answer (Q&A) of specialized knowledge in specific fields, search-based knowledge matching is typically implemented, including but not limited to question-answer pairs and knowledge graph retrieval. However, these methods often generate fixed, highly specialized answers that are difficult to understand. Furthermore, these search methods are limited and can easily fail to match questions or answers, reducing the accuracy and user-friendliness of the Q&A results.

[0005] Summary of the Invention

[0006] This application aims to solve at least one of the technical problems existing in the prior art or related art.

[0007] To this end, the first aspect of the present application is to propose a method for processing session information.

[0008] A second aspect of the present application is to provide a device for processing session information.

[0009] The third aspect of the present application is to propose a session information processing system.

[0010] The fourth aspect of this application proposes a sentence generation method.

[0011] The fifth aspect of the present application proposes a sentence generation device.

[0012] The sixth aspect of the present application proposes another sentence generation device.

[0013] A seventh aspect of the present application is to provide an electronic device.

[0014] The eighth aspect of the present application is to provide a readable storage medium.

[0015] A ninth aspect of the present application is to provide a cooking device.

[0016] In view of this, according to one aspect of the present application, a method for processing conversation information is proposed, the method comprising: receiving a first input from a user, the first input including first conversation information; in response to the first input, matching the first conversation information in a first database; if the match is successful, inputting second conversation information and first reply information corresponding to the first conversation information into a first model, so that the first model outputs second reply information based on the second conversation information and the first reply information; and displaying the second reply information; wherein the first database is a professional knowledge base in the professional field in which the first conversation information is located, and the first model is an intelligent dialogue model.

[0017] The conversation information processing method provided in this application is used to answer conversation information raised by users in professional fields such as cooking, and to provide users with accurate and easy-to-understand answer information, which can improve the accuracy of intelligent question and answer results and enhance the humanized features of intelligent question and answer.

[0018] Furthermore, the method for processing session information provided in the present application can be specifically executed by an electronic device that has data analysis and data processing functions.

[0019] Specifically, in the method for processing conversation information provided herein, when a user asks a question about a professional field, such as cooking, through an electronic device, the user can directly input the first conversation information through a first input into the electronic device. The electronic device receives and responds to the user's first input, comparing and matching the first conversation information input by the user with a first database, a professional knowledge base, for the professional field to which the first conversation information belongs. The first database stores multiple sets of conversation information and response information for the professional field to which the first conversation information belongs. On this basis, if the first database successfully matches the first conversation information input by the user, second conversation information and first response information corresponding to the first conversation information are obtained from the first database, where the first response information is more professional. Furthermore, the obtained first response information and second conversation information are input into an intelligent dialogue model, namely a first model, so that the first model optimizes the more professional first response information. Specifically, the first model integrates the first response information and the second conversation information, and based on the integrated information, the first response information is polished and corrected to obtain a less professional second response information that is easier for the user to understand, and the second response information is output and displayed to the user. In this way, by combining the professional knowledge base in the professional field and the intelligent dialogue model, intelligent answers can be given to questions raised by users. This can ensure the professionalism and thus the correctness of the question and answer results, while making it easier for users to understand, enhancing the humanization of the question and answer results and ensuring the effectiveness of the question and answer.

[0020] The method for processing session information described above in this application may also have the following additional technical features:

[0021] In some technical solutions, optionally, after matching the first session information in the first database, the method for processing the session information further includes: if the match fails, inputting the first session information into the first model so that the first model outputs third reply information based on the first session information; and displaying the third reply information.

[0022] In this technical solution, if the first database fails to match the first conversation information input by the user, the first conversation information is directly input into the intelligent dialogue model (i.e., the first model). This model then directly answers the first conversation information, generating third response information generated by the first model. This third response information is then output and displayed to the user. In this way, if the first conversation information fails to match the professional knowledge base in its field, the intelligent dialogue model provides a fallback answer to the first conversation information, ensuring efficient responses to user questions.

[0023] In some technical solutions, optionally, the first data includes multiple session information, and matching the first session information in the first database includes: encoding the first session information; performing similarity matching between the encoded first session information and multiple session information in the first database; if session information with the same semantics as the first session information exists in the first database, determining that the first session information successfully matches the first database; if any session information in the first database has a different semantics from the first session information, determining that the first session information fails to match the first database.

[0024] In this technical solution, after the electronic device receives the first session information input by the user, the electronic device will encode the first session information input by the user, and then perform semantic similarity matching on the above-mentioned first database and the encoded first session information. Furthermore, the above-mentioned first database stores multiple groups of session information and reply information in the professional field to which the above-mentioned first session information belongs. On this basis, if there is session information with the same or highly similar semantics as the first session information in the above-mentioned first database, the electronic device determines that the first database and the above-mentioned first session information are successfully matched. If there is no session information with the same or highly similar semantics as the first session information in the above-mentioned first database, the electronic device determines that the first database and the above-mentioned first session information have failed to match. In this way, the accuracy of the similarity matching results between the first database and the first session information is guaranteed, thereby ensuring the accuracy of the subsequent intelligent question and answer results.

[0025] In some technical solutions, optionally, before encoding the first session information, the method for processing session information further includes: preprocessing the first session information to obtain preprocessed first session information; wherein the preprocessing includes at least one of the following: extracting useful features from the first session information and filtering out noise information from the first session information.

[0026] In this technical solution, before the electronic device encodes the first conversation information input by the user, it also pre-processes the first conversation information to clean it, remove interference information, extract key feature information from the first conversation information, and thus produce semantically accurate first conversation information. This ensures the accuracy of the first conversation information involved in similarity matching, thereby ensuring the accuracy of the similarity matching results between the first database and the first conversation information, and ensuring the accuracy of subsequent intelligent question-and-answer results.

[0027] In some technical solutions, optionally, before encoding the first session information, the session information processing method further includes: if the language type of the first session information does not match the language type of the first database, inputting the first session information into a first model so that the first model converts the language type of the first session information into the language type of the first database; and displaying the second reply information includes: displaying the second reply information according to the language type of the first session information.

[0028] In this technical solution, after the electronic device receives the first conversation information input by the user and before the electronic device encodes the first conversation information, the electronic device also compares the language of the first database and the language of the first conversation information. On this basis, if the language of the first database and the first conversation information are different, the electronic device outputs the first conversation information input by the user to the first model, and uses the first model to convert the language of the first conversation information to obtain the first conversation information with the same language as the language of the first database. Then, based on the matching result between the converted first conversation information and the first database, the electronic device answers the user's question. In particular, if the language of the first database and the first conversation information are different, after the electronic device answers the first conversation information, the electronic device outputs and displays a second reply message in the same language as the language of the first conversation information input by the user, to ensure the consistency of the language of the input and output information, which is convenient for the user to understand. In this way, with the help of the translation capability of the first model, multilingual intelligent question and answer dialogue is realized, the application scope of intelligent question and answer is expanded, and the question and answer effect is guaranteed.

[0029] In some technical solutions, optionally, the encoding process includes at least one of the following: vector encoding process, cosine encoding process.

[0030] In this technical solution, when the electronic device encodes the first session information input by the user, it may specifically perform cosine encoding or vector encoding on the first session information to obtain corresponding first encoded information.

[0031] In some technical solutions, optionally, the similarity matching algorithm includes at least one of the following: a similarity matching algorithm based on Euclidean distance, a similarity matching algorithm based on cosine distance, and a similarity matching algorithm based on a deep learning model.

[0032] In this technical solution, in the process of the electronic device performing semantic similarity matching on the above-mentioned first database and the first conversation information input by the user, a similarity matching algorithm based on a deep learning model, a similarity matching algorithm based on cosine distance, or a similarity matching algorithm based on Euclidean distance can be specifically adopted.

[0033] According to a second aspect of the present application, a device for processing conversation information is provided, the device comprising: a receiving unit for receiving a first input from a user, the first input comprising first conversation information; a processing unit for matching the first conversation information in a first database in response to the first input; if the match is successful, inputting second conversation information and first reply information corresponding to the first conversation information into a first model, so that the first model outputs second reply information based on the second conversation information and the first reply information; and a display unit for displaying the second reply information; wherein the first database is a professional knowledge base for the professional field in which the first conversation information is located, and the first model is an intelligent dialogue model.

[0034] The conversation information processing device provided in this application is used to answer conversation information raised by users in professional fields such as cooking, and feedback accurate and easy-to-understand reply information to users, which can improve the correctness of intelligent question and answer results and enhance the humanized features of intelligent question and answer.

[0035] Specifically, the session information processing device provided in the present application includes a receiving unit and a processing unit. When a user asks questions about a professional field, such as the cooking field, through an electronic device, the user can directly input the first session information through a first input to the electronic device. The receiving unit receives the first input of the user, and the processing unit responds to the first input to obtain the first session information input by the user, and compares and matches the first session information with the professional knowledge base of the professional field to which it belongs, that is, the first database. The above-mentioned first database stores multiple groups of session information and reply information of the professional field to which the above-mentioned first session information belongs. On this basis, when the above-mentioned first database successfully matches the first session information input by the user, the processing unit obtains the second session information and the first reply information corresponding to the above-mentioned first session information in the first database, and the first reply information is more professional. Furthermore, the processing unit inputs the acquired first reply information and second conversation information into the intelligent dialogue model, i.e., the first model, to optimize the more professional first reply information through the first model. Specifically, the first model integrates the first reply information and the second conversation information, and based on the integrated information, polishes and corrects the first reply information, thereby obtaining less professional second reply information that is easier for the user to understand, and outputs the second reply information for display to the user. In this way, by combining the professional knowledge base of the professional field and the intelligent dialogue model, intelligent answers to questions raised by users can be provided, ensuring the professionalism and thus the correctness of the question and answer results while being easier for users to understand, enhancing the humanization of the question and answer results, and ensuring the question and answer effect.

[0036] According to the third aspect of the present application, a system for processing conversation information is proposed, comprising: a first model, which is an intelligent dialogue model; a first database, which stores multiple groups of conversation information and reply information; and a processor, which is used to implement the steps of the conversation information processing method in any of the above technical solutions.

[0037] The third aspect of the present application provides a system for processing conversation information, including a first model, a first database, and a processor. The first model is an intelligent dialogue model, and the first database is a professional knowledge base. The first database stores multiple sets of conversation information and response information in a professional field, such as cooking.

[0038] Furthermore, the processor is configured to implement the steps of the method for processing session information in any of the above technical solutions. Therefore, the session information processing system proposed in the third aspect of the present application has all the beneficial effects of the method for processing session information in any of the above technical solutions in the first aspect, and will not be described in detail here.

[0039] According to a fourth aspect of the present application, a sentence generation method is proposed, including: upon receiving a first question sentence, identifying intent information in the first question sentence; determining a target question sentence that matches the first question sentence based on the intent information; and generating an answer sentence based on a target corpus of at least two corpora, wherein the target corpus matches the sentence category of the target question sentence.

[0040] The technical solution of the present application proposes a sentence generation method, which is applied to a question-answering system. The question-answering system can answer the first question sentence input by the user into the question-answering system through the sentence generation method. The question-answering system can determine the target question sentence that matches the first question sentence by means of intent recognition, and find the corresponding target corpus based on the target question sentence, and generate an answer sentence through the target corpus in the target corpus. The present application realizes the distribution of different types of questions by first performing intent recognition and classification on the first question sentence, that is, generating corresponding answer sentences through the target corpus in different target corpora. Different types of question sentences can be answered through only one intent recognition, which reduces the waste of computing power and improves the efficiency of generating answer sentences.

[0041] In this technical solution, the question-answering system includes an intent matching module that can identify the intent information of a first question statement and, based on the intent information, find a target question statement corresponding to the first question statement. The target question statement found is a preset question statement, and the semantics of the target question statement are similar to or identical to those of the first question statement. That is, through the identified intent information, the first question statement input by the user can be converted into a standardized target question statement.

[0042] In this technical solution, the question-answering system can find a matching target corpus in different corpora based on the sentence category. Sentence categories include but are not limited to structured question sentences and preset question sentences. The question-answering system determines the target question sentence and the sentence category corresponding to the target question sentence through intent recognition, so that it can accurately select the corresponding target corpus. This eliminates the need to set up recognition modules that adapt to different sentence categories in the question-answering system, reduces the computing power loss of the question-answering system, and improves processing efficiency.

[0043] Structured question statements are typically simple, real-time questions, stored in a structured manner. Examples include recipe ingredients and quantities, cooking time, cooking temperature, and cooking steps. Preset question statements are typically complex questions, typically stored in a question-and-answer format. This means that the corresponding standard answer is found through the preset question statement and output as the answer statement.

[0044] In the technical solution of the present application, after receiving the first question sentence, the question-answering system can determine the standardized target question sentence corresponding to the first question sentence through intent recognition, and based on the sentence category of the target question sentence, find the corresponding target corpus, and generate an answer sentence through the target corpus in the target corpus. There is no need to repeatedly process the first question sentence according to different sentence categories, which reduces the energy consumption of the question-answering system and improves the efficiency of the question-answering system in outputting answer sentences.

[0045] In some technical solutions, optionally, determining a target question statement that matches the first question statement based on the intent information includes: obtaining at least two second question statements based on the intent information, and the at least two second question statements are associated with the intent information; and screening the target question statement among the at least two second question statements.

[0046] This technical solution proposes that intention information obtained through intention recognition can be used to obtain multiple second question statements with semantics similar to the first question statement, and then the target question statement matching the first question statement is obtained by screening the multiple second question statements, thereby improving the accuracy of the screened target question statement.

[0047] In the technical solution of the present application, an intention matching module is configured in the question-answering system. The intention matching module can initially recall candidate similar second question statements. After obtaining multiple second question statements, the closest target question statement can be obtained by comparing the multiple second question statements with the first question statement respectively.

[0048] In the technical solution of the present application, the question-answering system can perform two-level screening through intent information, that is, first find multiple semantically similar second question sentences, and then screen the target question sentences that match the semantics of the first question sentences among the multiple second question sentences, thereby improving the accuracy of obtaining the matching target question sentences.

[0049] The sentence generation method of the present application may also have the following additional technical features:

[0050] In some technical solutions, optionally, screening the target question statement from the at least two second question statements includes: obtaining the semantic similarity between the first question statement and each of the at least two second question statements; and screening the target question statement from the at least two second question statements based on the semantic similarity.

[0051] This technical solution proposes a process of screening a target question sentence from multiple second question sentences by using semantic similarity.

[0052] In the technical solution of the present application, the semantic similarity between the first question statement and each second question statement is calculated respectively, multiple second question statements are sorted according to the semantic similarity, and the second question statement with the highest semantic similarity is used as the target question statement, thereby improving the matching degree between the screened target question statement and the first question statement.

[0053] Among them, after obtaining multiple second question statements, the question answering system calculates the semantic similarity between the multiple second question statements and the first question statement, and then selects the second question statement with the highest semantic similarity value as the matched target question statement.

[0054] In the technical solution of the present application, after the question-answering system can find multiple second question statements through intent information, the question-answering system can accurately calculate multiple semantic similarities between the multiple second question statements and the first question statement, and based on the semantic similarities, find the target question statement among the multiple second question statements, further improving the matching degree between the target question statement and the first question statement.

[0055] In some technical solutions, optionally, obtaining the semantic similarity between the first question statement and each of the at least two second question statements includes: obtaining a first semantic vector of the first question statement and a second semantic vector of each second question statement; and calculating the semantic similarity through the first semantic vector and the second semantic vector.

[0056] This technical solution proposes a solution of converting multiple second question statements and the first question statement into semantic vectors, and calculating the semantic similarity between each second question statement and the first question statement based on the semantic vectors.

[0057] In the technical solution of the present application, first semantic vectors and second semantic vectors corresponding to the first question statement and the second question statement are obtained, respectively. The number of second semantic vectors is the same as the number of second question statements, and the second semantic vectors have a one-to-one correspondence with the second question statements. By calculating similarity between the first semantic vector and the second semantic vector, multiple semantic similarities can be obtained, and the semantic similarities have a one-to-one correspondence with the second semantic vectors.

[0058] In the technical solution of the present application, by converting both the first question statement and the second question statement into high-dimensional semantic vectors and screening the corresponding target question statement by calculating the similarity of the high-dimensional semantic vectors, the accuracy of determining the target question statement can be improved.

[0059] In some technical solutions, optionally, obtaining the semantic similarity between the first question statement and each of the at least two second question statements includes: inputting the first question statement and each second question statement into a pre-trained model to determine the semantic similarity.

[0060] This technical solution proposes a process of inputting all the first question statements and the second question statements into a pre-training model, and calculating multiple semantic similarities between multiple second question statements and the first question area through the training model.

[0061] In the technical solution of the present application, a pre-training model is set in the question-answering system, and when multiple second question statements are obtained, the multiple second question statements and the first question statement are input into the pre-training model. The pre-training model can accurately calculate multiple semantic similarities between the multiple second question statements and the first question statement.

[0062] In some technical solutions, optionally, before generating an answer sentence based on a target corpus from at least two corpora, the process includes: selecting a target corpus from at least two corpora based on identification information in a target question sentence, the identification information corresponding to a sentence category; wherein the sentence category includes at least one of the following: a preset question category and a structured question category.

[0063] This technical solution proposes that the question-answering system can filter out a target corpus from at least two corpora based on the identification information corresponding to the target question sentence, that is, it ensures that the question-answering system can determine the target corpus that matches the target question sentence corresponding to the first question sentence after determining it.

[0064] In the technical solution of the present application, the target question sentence is a question sentence preset in the question-answering system, and each target question sentence is provided with corresponding identification information. After the question-answering system determines the target question sentence corresponding to the first question sentence, it can determine the sentence category of the target question sentence by reading the identification information of the target question sentence, thereby finding the corresponding target corpus.

[0065] In the technical solution of the present application, the number of corpora configured in the question-answering system matches the number of sentence categories and the number of identification information, and the corpora, sentence categories and identification information correspond one to one.

[0066] In the technical solution of the present application, the sentence category includes at least one of a preset question category and a structured question category. Structured question category sentences are typically simple real-time questions, which are stored in a structured manner. Preset question category sentences are typically complex questions, typically stored in a question-answer format. That is, the corresponding standard answer is found through the preset question category sentence, and the standard answer is output as the answer sentence.

[0067] In the technical solution of the present application, by configuring corresponding identification information in the target question sentence pre-stored in the question-answering system, it is ensured that after the question-answering system obtains the target question-answer sentence that matches the first question-answer sentence, it can quickly determine the sentence category corresponding to the target question-answer sentence, and determine the target corpus among multiple corpora accordingly.

[0068] In some technical solutions, optionally, the sentence category is a structured question category; generating an answer sentence based on a target corpus in at least two corpora includes: determining a target query sentence based on a target question sentence; searching a target corpus in a target corpus through the target query sentence; and generating an answer sentence based on the target corpus.

[0069] This technical solution proposes a specific process for a question-answering system to generate an answer statement when the target question statement is a structured question-type statement.

[0070] In the technical solution of the present application, when the sentence category is a structured question, a target query sentence matching the target question sentence is searched. The target query sentence is used to search for the corresponding target corpus in the target corpus corresponding to the structured question class. After the corresponding target corpus is found, the corresponding answer sentence is generated through the target corpus.

[0071] The target query statement is a structured query code.

[0072] Specifically, when the target question statement is a structured question-type statement, the template of the corresponding target query statement can be found through the target question statement. By extracting the entity nouns in the target question statement and inserting the entity nouns into the template, the target query statement can be obtained. Through the target query statement, the target corpus can be queried in the target corpus, and the answer statement can be output based on the target corpus.

[0073] In the technical solution of the present application, when the target question statement is a structured question-type statement, the question-answering system can first determine the corresponding target query statement based on the target question statement, then find the target corpus in the corresponding target corpus through the target query statement, and then generate the corresponding answer statement based on the target corpus to ensure the degree of matching between the answer statement and the target question statement.

[0074] In some technical solutions, optionally, a target query statement is determined based on a target question statement, including: extracting noun vocabulary and other vocabulary in the target question statement; determining a matching query statement template based on the other vocabulary; and inputting the noun vocabulary into the query statement template to generate a target query statement.

[0075] This technical solution proposes a process of searching for a query statement template of a target query statement when the target question statement is a structured question statement, and determining the target query statement based on the query statement template.

[0076] In the technical solution of the present application, the target query statement corresponding to the structured question statement is also a structured query statement, that is, different target query statements correspond to different query statement templates. The query statement template can be determined by parsing other words in the target question statement, and then the noun words in the target question statement are inserted into the query statement template to determine the target query statement.

[0077] Among them, noun vocabulary is the entity vocabulary in the target question sentence, and other vocabulary is the non-entity vocabulary in the target sentence, that is, all vocabulary except noun vocabulary.

[0078] In the technical solution of the present application, the question-answering system can find a query sentence template that matches the target question sentence by performing semantic analysis on other words.

[0079] In the technical solution of the present application, after determining the corresponding query sentence template, the noun vocabulary in the target question sentence is inserted into the query sentence template to obtain the corresponding target query sentence.

[0080] In the technical solution of the present application, by parsing the target question sentence of the structured question type, the noun vocabulary and other vocabulary therein can be obtained, and the corresponding query sentence template is queried through other vocabulary. Then, the noun vocabulary is inserted into the query sentence template to obtain the corresponding target query sentence, which improves the accuracy of the target query sentence and further improves the accuracy of the target corpus found, thereby improving the matching degree between the generated answer sentence and the first question sentence.

[0081] In some technical solutions, optionally, the sentence category is a preset question category; generating an answer sentence based on a target corpus in at least two corpora includes: searching for an answer sentence according to a target question sentence and a preset mapping relationship, where the preset mapping relationship is a correspondence between the target question sentence and the answer sentence in the target corpus.

[0082] This technical solution proposes a specific process for a question-answering system to generate an answer statement when the target question statement is a preset question-type statement.

[0083] In the technical solution of the present application, the question-answering system stores answer statements corresponding to target question statements of a preset question class, as well as a preset mapping relationship between the answer statements and the target question statements of the preset question class.

[0084] The target corpus corresponding to the target question sentence of the preset question class stores a preset mapping relationship between the answer sentence and the target question sentence of the preset question class, as well as the corresponding answer sentence. After determining the target question sentence of the preset question class, the corresponding answer sentence is found using the preset mapping relationship in the target corpus and the answer sentence is output.

[0085] In the technical solution of the present application, when the target question statement is a preset question-type statement, the corresponding answer statement can be automatically found in the target corpus through the preset mapping relationship, thereby ensuring the relevance and matching degree between the answer statement and the first question statement.

[0086] In some technical solutions, optionally, before identifying the intention information in the first question statement, the method further includes: upon receiving the first question statement, updating the first question statement according to vocabulary in a preset vocabulary library.

[0087] This technical solution proposes that after receiving the first question statement, the first question statement is updated by using vocabulary in a preset vocabulary library, which can improve the accuracy and search efficiency of the question-answering system in finding the target question statement through semantic information.

[0088] In an embodiment of the present application, the preset vocabulary library is a vocabulary library pre-stored in the question-answering system. Updating the vocabulary in the first question sentence by using the vocabulary in the preset vocabulary library can improve the recognition efficiency and accuracy of the semantic recognition of the first question sentence by the question-answering system.

[0089] In some technical solutions, optionally, when the first question statement is received, the first question statement is updated based on the vocabulary in the preset vocabulary library, including: extracting the entity nouns in the first question statement; based on the entity nouns, searching for corresponding abstract nouns in the preset vocabulary library; replacing the entity nouns in the first question statement with abstract nouns to update the first question statement.

[0090] This technical solution proposes a specific process of replacing entity nouns in the first question sentence with abstract nouns in a preset vocabulary library.

[0091] Specifically, entity nouns are nouns with a smaller reference scope in the first question sentence, and abstract nouns are nouns with a larger reference scope in the preset vocabulary library. There is a semantic inclusion relationship between the abstract nouns found through entity nouns.

[0092] In the technical solution of the present application, different entity nouns may correspond to the same abstract noun in the preset vocabulary library.

[0093] In the technical solution of the present application, by replacing the entity nouns in the first question statement with abstract nouns in a preset vocabulary library after receiving the first question statement, the computing power required to extract the intent information of the first question statement can be reduced, and the efficiency of obtaining the intent information corresponding to the first question statement can be improved.

[0094] According to the fifth aspect of the present application, a sentence generation device is proposed, including: an identification module for identifying intent information in the first question sentence when a first question sentence is received; a determination module for determining a target question sentence that matches the first question sentence based on the intent information; and a generation module for generating an answer sentence based on a target corpus of at least two corpora, wherein the target corpus matches the sentence category of the target question sentence.

[0095] The technical solution of the present application proposes a sentence generation device, which is applied to a question-answering system. The question-answering system can answer the first question sentence input by the user into the question-answering system through the sentence generation device. The question-answering system can determine the target question sentence that matches the first question sentence by means of intent recognition, and find the corresponding target corpus based on the target question sentence, and generate an answer sentence through the target corpus in the target corpus. The present application realizes the distribution of different types of questions by first performing intent recognition and classification on the first question sentence, that is, generating corresponding answer sentences through the target corpus in different target corpora. Different types of question sentences can be answered through only one intent recognition, which reduces the waste of computing power and improves the efficiency of generating answer sentences.

[0096] In this technical solution, the question-answering system includes an intent matching module that can identify the intent information of a first question statement and, based on the intent information, find a target question statement corresponding to the first question statement. The target question statement found is a preset question statement, and the semantics of the target question statement are similar to or identical to those of the first question statement. That is, through the identified intent information, the first question statement input by the user can be converted into a standardized target question statement.

[0097] In this technical solution, the question-answering system can find a matching target corpus in different corpora based on the sentence category. Sentence categories include but are not limited to structured question sentences and preset question sentences. The question-answering system determines the target question sentence and the sentence category corresponding to the target question sentence through intent recognition, so that it can accurately select the corresponding target corpus. This eliminates the need to set up recognition modules that adapt to different sentence categories in the question-answering system, reduces the computing power loss of the question-answering system, and improves processing efficiency.

[0098] Structured question statements are typically simple, real-time questions, stored in a structured manner. Examples include recipe ingredients and quantities, cooking time, cooking temperature, and cooking steps. Preset question statements are typically complex questions, typically stored in a question-and-answer format. This means that the corresponding standard answer is found through the preset question statement and output as the answer statement.

[0099] In the technical solution of the present application, after receiving the first question sentence, the question-answering system can determine the standardized target question sentence corresponding to the first question sentence through intent recognition, and based on the sentence category of the target question sentence, find the corresponding target corpus, and generate an answer sentence through the target corpus in the target corpus. There is no need to repeatedly process the first question sentence according to different sentence categories, which reduces the energy consumption of the question-answering system and improves the efficiency of the question-answering system in outputting answer sentences.

[0100] According to the sixth aspect of the present application, another sentence generation device is proposed, including: a memory, in which a program or instruction is stored; a processor, which executes the program or instruction stored in the memory to implement the steps of the sentence generation method in any technical solution in the fourth aspect, thereby having all the beneficial technical effects of the sentence generation method in any technical solution in the above-mentioned fourth aspect, and will not be elaborated on here.

[0101] According to the seventh aspect of the present application, an electronic device is proposed, comprising a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method for processing session information as described in any of the above technical solutions are implemented, or when the program or instruction is executed by the processor, the steps of the method for generating sentences as described in any of the above technical solutions are implemented. Therefore, the electronic device proposed in the seventh aspect of the present application has all the beneficial effects of the method for processing session information as described in any of the above technical solutions in the first aspect, or the electronic device proposed in the seventh aspect of the present application has all the beneficial effects of the method for generating sentences as described in any of the above technical solutions in the fourth aspect, which will not be repeated here.

[0102] According to the eighth aspect of the present application, a readable storage medium is provided, on which a program or instruction is stored. When executed by a processor, the program or instruction implements the method for processing session information as described in any of the above-mentioned technical solutions, or when executed by a processor, the program or instruction implements the method for generating statements as described in any of the above-mentioned technical solutions. Therefore, the readable storage medium provided in the eighth aspect of the present application has all the beneficial effects of the method for processing session information as described in any of the above-mentioned technical solutions in the first aspect, or the readable storage medium provided in the eighth aspect of the present application has all the beneficial effects of the method for generating statements as described in any of the above-mentioned technical solutions in the fourth aspect, and these will not be repeated here.

[0103] According to the ninth aspect of the present application, a cooking device is proposed, comprising: a sentence generating device as defined in the fifth or sixth aspect above, and / or a readable storage medium as defined in the eighth aspect above, and thus having all the beneficial technical effects of the sentence generating device in the fifth or sixth aspect above, and / or the readable storage medium defined in the eighth aspect above, and no further details will be given here.

[0104] Additional aspects and advantages of the present application will become apparent in the following description or may be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0105] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0106] FIG1 shows a flow chart of a method for processing session information according to an embodiment of the present application;

[0107] FIG2 shows a second flow chart of a method for processing session information according to an embodiment of the present application;

[0108] FIG3 shows a third flow chart of a method for processing session information according to an embodiment of the present application;

[0109] FIG4 shows a block diagram of a device for processing session information according to an embodiment of the present application;

[0110] FIG5 shows a block diagram of a system for processing session information according to an embodiment of the present application;

[0111] FIG6 shows one of the schematic flow charts of the sentence generation method according to an embodiment of the present application;

[0112] FIG7 shows a second schematic flow chart of the sentence generation method according to an embodiment of the present application;

[0113] FIG8 shows a third flow chart of the sentence generation method according to an embodiment of the present application;

[0114] FIG9 shows a fourth flow chart of the sentence generation method according to an embodiment of the present application;

[0115] FIG10 shows a fifth flow chart of the sentence generation method according to an embodiment of the present application;

[0116] FIG11 shows a structural block diagram of a sentence generating apparatus according to an embodiment of the present application;

[0117] FIG12 shows a second structural block diagram of the sentence generation device according to an embodiment of the present application;

[0118] FIG13 shows a structural block diagram of an electronic device according to an embodiment of the present application;

[0119] FIG14 shows a structural block diagram of a cooking device according to an embodiment of the present application. DETAILED DESCRIPTION

[0120] In order to more clearly understand the above-mentioned objects, features and advantages of the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features therein can be combined with each other in the absence of conflict.

[0121] In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0122] 1 to 14 , the conversation information processing method, device, and system, and the statement generation method and device provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0123] In one embodiment of the present application, as shown in FIG1 , the method for processing session information may specifically include the following steps 102 to 108:

[0124] Step 102, receiving a first input from a user;

[0125] Step 104, in response to the first input, matching the first session information in the first database;

[0126] Step 106: If the match is successful, the second session information and the first reply information corresponding to the first session information are input into the first model, so that the first model outputs the second reply information based on the second session information and the first reply information.

[0127] Step 108, displaying the second reply information;

[0128] Among them, the above-mentioned first model is an intelligent dialogue model, and the above-mentioned first database is a professional knowledge base of the professional field to which the above-mentioned first conversation information belongs, such as the cooking field.

[0129] The conversation information processing method provided in this application is used to answer conversation information raised by users in professional fields such as cooking, and to provide users with accurate and easy-to-understand answer information, which can improve the accuracy of intelligent question and answer results and enhance the humanized features of intelligent question and answer.

[0130] Furthermore, the method for processing session information provided in the present application can be specifically executed by an electronic device that has data analysis and data processing functions.

[0131] In actual application, the above-mentioned electronic devices can specifically be smart devices such as smart phones, smart watches, tablet computers and laptops, and are not specifically limited here.

[0132] Specifically, in the method for processing conversation information provided herein, when a user asks a question about a professional field, such as cooking, through an electronic device, the user can directly input the first conversation information through a first input into the electronic device. The electronic device receives and responds to the user's first input, comparing and matching the first conversation information input by the user with a first database, a professional knowledge base, for the professional field to which the first conversation information belongs. The first database stores multiple sets of conversation information and response information for the professional field to which the first conversation information belongs. On this basis, if the first database successfully matches the first conversation information input by the user, second conversation information and first response information corresponding to the first conversation information are obtained from the first database, where the first response information is more professional. Furthermore, the obtained first response information and second conversation information are input into an intelligent dialogue model, namely a first model, so that the first model optimizes the more professional first response information. Specifically, the first model integrates the first response information and the second conversation information, and based on the integrated information, the first response information is polished and corrected to obtain a less professional second response information that is easier for the user to understand, and the second response information is output and displayed to the user. In this way, by combining the professional knowledge base in the professional field and the intelligent dialogue model, intelligent answers can be given to questions raised by users. This can ensure the professionalism and thus the correctness of the question and answer results, while making it easier for users to understand, enhancing the humanization of the question and answer results and ensuring the effectiveness of the question and answer.

[0133] Among them, the above-mentioned first input can specifically be the user's touch input to the question and answer interface in the electronic device. The touch input can specifically be single-click input, double-click input, sliding input, input along a preset trajectory, etc. Technical personnel in this field can set the specific form of the above-mentioned first input according to actual conditions, and no specific restrictions are made here.

[0134] In actual application, an information input control may be displayed in the above-mentioned question-and-answer interface. The above-mentioned first input may specifically be the user's touch input to the information input control. The electronic device responds to the touch input and outputs the corresponding second reply information for the first conversation information proposed by the user.

[0135] The information input control may be a text input control or a voice input control. Furthermore, the first conversation information may be a text message or a voice message, which is not specifically limited here.

[0136] On this basis, when the electronic device answers the first conversation message input by the user and outputs the second reply message, if the first conversation message is textual, the electronic device answers the first conversation message directly based on the textual information. If the first conversation message is voice information, the electronic device converts the voice information into textual information and then answers the first conversation message based on the textual information.

[0137] Furthermore, the first database stores multiple sets of conversation information and response information in the professional field to which the first conversation information belongs. During the comparison and matching process between the first database and the first conversation information input by the user, the first conversation information input by the user is specifically compared with the multiple conversation information stored in the first database for semantic similarity. Based on this comparison, if conversation information with the same or highly similar semantics as the first conversation information exists in the first database, the first database is determined to have successfully matched the first conversation information. If no conversation information with the same or highly similar semantics as the first conversation information exists in the first database, the first database is determined to have failed to match the first conversation information.

[0138] In addition, in actual application, the first database may be a database pre-stored in a storage area of ​​the electronic device, or a shared database of a third-party software or website, which is not specifically limited here.

[0139] In some embodiments of the present application, optionally, after step 104, the method for processing session information may further include the following steps 110 and 112:

[0140] Step 110: If the match fails, input the first session information into the first model, so that the first model can output third reply information based on the first session information.

[0141] Step 112: Display the third reply information.

[0142] In this embodiment, after the electronic device receives first conversation information input by the user and, in response to the user's first input, compares and matches the first conversation information with a first database, a professional knowledge base in the professional field to which the first conversation information belongs, if the first database fails to match the first conversation information input by the user, the electronic device directly inputs the first conversation information input by the user into the first model, so that the first model directly answers the first conversation information input by the user, obtains third reply information generated by the first model, and outputs and displays the third reply information to the user. In this way, if the first conversation information fails to match the professional knowledge base in the professional field to which it belongs, the intelligent conversation model provides a fallback answer to the first conversation information, ensuring efficient response to user questions.

[0143] Among them, the output mode of the above-mentioned first model is a streaming output mode to ensure the real-time feedback of the question and answer results and improve the question and answer effect.

[0144] In some embodiments of the present application, the first data may optionally include multiple session information. Based on this, the step of matching the first session information in the first database may specifically include the following steps 104a to 104d:

[0145] Step 104a, encoding the first session information;

[0146] Step 104b, performing similarity matching on the first database and the encoded first session information;

[0147] Step 104c: if there is session information in the first database that has the same semantics as the first session information, determine that the first database successfully matches the first session information;

[0148] Step 104d: If there is no session information in the first database that has the same semantics as the first session information, it is determined that the first database fails to match the first session information.

[0149] In this embodiment, after the electronic device receives the first conversation information input by the user, the electronic device also encodes the first conversation information input by the user and then performs semantic similarity matching on the first database and the encoded first conversation information. Furthermore, the first database stores multiple sets of conversation information and response information in the professional field to which the first conversation information belongs. On this basis, if there is conversation information in the first database that is semantically identical or highly similar to the first conversation information, the electronic device determines that the first database successfully matches the first conversation information and inputs the second conversation information and the first response information corresponding to the first conversation information in the first database into the intelligent dialogue model, namely the first model, so that after optimizing the highly professional first response information through the first model, the second response information is presented to the user in an easier-to-understand manner. If there is no conversation information in the first database that is semantically identical or highly similar to the first conversation information, the electronic device determines that the first database fails to match the first conversation information and inputs the first conversation information input by the user into the intelligent dialogue model, namely the first model, so that the first model directly answers the first conversation information input by the user and displays the corresponding third response information to the user. In this way, the accuracy of the similarity matching results between the first database and the first conversation information is guaranteed, thereby ensuring the accuracy of subsequent intelligent question and answer results.

[0150] In the process of encoding the first session information input by the user, the first session information may be subjected to cosine encoding or vector encoding to obtain corresponding first encoded information. In actual applications, those skilled in the art may select the specific encoding method based on actual circumstances, and no specific limitation is imposed herein.

[0151] Furthermore, when performing semantic similarity matching on the first database and the encoded first conversation information, a similarity matching algorithm based on Euclidean distance, cosine distance, or deep learning models such as LSTM (Long Short-Term Memory) and BERT (Bidirectional Encoder Representation from Transformers) models can be used. In actual applications, the specific type of similarity matching algorithm can be selected by those skilled in the art based on actual circumstances and is not specifically limited herein.

[0152] Furthermore, the above encoding processing method corresponds to the similarity matching algorithm to ensure the accuracy of the similarity matching result.

[0153] Furthermore, the first conversation information may be a text message or a voice message.

[0154] On this basis, when the electronic device performs semantic similarity matching on the first database and the encoded first conversation information, if the first conversation information is text information, the electronic device directly performs semantic similarity matching on the first database and the first conversation information based on the text information. If the first conversation information is voice information, the electronic device converts the voice information into text information and then performs semantic similarity matching on the first database and the first conversation information based on the text information.

[0155] In some embodiments of the present application, optionally, before step 104a, the method for processing session information may further include the following step 118:

[0156] Step 118: pre-process the first session information to obtain pre-processed first session information;

[0157] The preprocessing includes at least one of the following: filtering out noise information in the first session information, and extracting useful features in the first session information.

[0158] In this embodiment, before the electronic device encodes the first conversation information input by the user, it also pre-processes the first conversation information input by the user to clean the first conversation information, remove interference information from the first conversation information, extract key feature information from the first conversation information, and thus generate semantically accurate first conversation information. This ensures the accuracy of the first conversation information involved in similarity matching, thereby ensuring the accuracy of the similarity matching results between the first database and the first conversation information, and ensuring the accuracy of subsequent intelligent question and answer results.

[0159] In actual application, the preprocessing may specifically include: extracting useful features from the first session information to obtain key feature information from the first session information.

[0160] Furthermore, in actual application, the above preprocessing may further specifically include: filtering out noise information in the first session information to remove interference information in the first session information.

[0161] In actual application, those skilled in the art may select the specific method of the above pretreatment according to actual conditions, and no specific limitation is imposed here.

[0162] In some embodiments of the present application, optionally, before step 104a, the method for processing session information may further include steps 120 and 122. On this basis, step 108 may further include step 108a:

[0163] Step 120 , if the language types of the first database and the first conversation information are different, input the first conversation information into the first model;

[0164] Step 122: convert the language of the first conversation information using the first model so that the language of the first database and the first conversation information are the same;

[0165] Step 108a: Display the second reply information according to the language type of the first conversation information.

[0166] In this embodiment, after the electronic device receives the first conversation information input by the user and before it encodes the first conversation information, it compares the language of the first database and the language of the first conversation information. Based on this comparison, if the language of the first database and the first conversation information are the same, the electronic device directly answers the user's question based on the matching result between the first database and the first conversation information. If the language of the first database and the first conversation information are different, the electronic device outputs the first conversation information input by the user to the first model, where the first model converts the language of the first conversation information to obtain first conversation information in the same language as the language of the first database. The electronic device then answers the user's question based on the matching result between the converted first conversation information and the first database. If the language of the first database and the first conversation information are different, after the electronic device answers the first conversation information, it outputs and displays a second reply message in the same language as the first conversation information input by the user, ensuring consistency in the language of the input and output information for easier understanding by the user. In this way, with the help of the translation capability of the first model, multi-language intelligent question-answering dialogues are realized, the application scope of intelligent question-answering is expanded, and the question-answering effect is guaranteed.

[0167] In some embodiments of the present application, optionally, the above encoding processing includes at least one of the following: cosine encoding processing, vector encoding processing.

[0168] In this embodiment, when the electronic device encodes the first session information input by the user, the electronic device may specifically perform cosine encoding or vector encoding on the first session information to obtain corresponding first encoded information.

[0169] In actual application, those skilled in the art may select the specific method of the above encoding processing according to actual conditions, and no specific limitation is made here.

[0170] In some embodiments of the present application, optionally, the above-mentioned similarity matching algorithm includes at least one of the following: a similarity matching algorithm based on a deep learning model, a similarity matching algorithm based on cosine distance, and a similarity matching algorithm based on Euclidean distance.

[0171] In this embodiment, when the electronic device performs semantic similarity matching on the first database and the first conversation information input by the user, a similarity matching algorithm based on a deep learning model such as an LSTM or BERT model, a similarity matching algorithm based on a cosine distance, or a similarity matching algorithm based on a Euclidean distance may be specifically adopted.

[0172] In actual application, those skilled in the art may select the specific type of the similarity matching algorithm according to actual conditions, and no specific limitation is imposed herein.

[0173] Furthermore, the above similarity matching algorithm corresponds to the above encoding processing method to ensure the accuracy of the similarity matching result.

[0174] In some embodiments of the present application, optionally, after step 108, the method for processing session information may further include the following steps 126 and 128:

[0175] Step 126, obtaining user feedback information;

[0176] Step 128: updating the database information of the first database according to the feedback information;

[0177] The feedback information is associated with the second reply information.

[0178] In this embodiment, after the electronic device responds to the first conversation information input by the user, the user can provide feedback on the second response information output by the electronic device through touch input on the question-and-answer interface of the electronic device. On this basis, the electronic device receives and responds to the user's touch input on the question-and-answer interface and optimizes the database content of the first database based on the feedback information entered by the user in the question-and-answer interface, that is, updates the professional information data stored in the first database. In this way, based on the user's evaluation of the second response information, the database content of the first database is updated and optimized, which increases the data range of the first database, thereby further improving the accuracy of subsequent responses to user questions.

[0179] Specifically, after the electronic device responds to the first conversation information input by the user, the electronic device may display a target control within the question-and-answer interface, which is used to evaluate the second reply information output by the electronic device. Based on this, the electronic device may receive and respond to the user's touch input to the target control, and based on the evaluation information fed back by the user to the electronic device, update the database content of the first database to optimize the database content corresponding to the first conversation information in the first database, or add database content associated with the first conversation information to the first database.

[0180] In some embodiments of the present application, optionally, as shown in FIG2 , the method for processing the session information may specifically include the following steps 202 to 214:

[0181] Step 202, receiving user questions;

[0182] Step 204, encode the user question;

[0183] Step 206, matching the user question with the professional database;

[0184] Step 208, determine whether the matching result is: successful matching, if so, execute step 210, if not, execute step 212;

[0185] Step 210: Send the professional database matching questions and answers together into the big model to optimize the answers;

[0186] Step 212: Send the user's question to the big model to directly generate the answer;

[0187] Step 214: output the answer.

[0188] Among them, the large model in step 210 and step 212 is equivalent to the above-mentioned first model, the professional library in step 206 and step 210 is equivalent to the above-mentioned first database, and the professional library matching questions and answers in step 210 is equivalent to the above-mentioned second conversation information and first reply information.

[0189] In this embodiment, specifically, after a user question is entered, the electronic device first encodes the user question. Prior to encoding, the user question needs to be cleaned, which may include steps such as screening for useful features and removing noise. Furthermore, encoding methods for the user question include, but are not limited to, vector encoding and cosine encoding. Based on this, the encoded user question is semantically matched against a professional database. Similarity matching algorithms include those based on Euclidean distance, cosine distance, or deep learning-based LSTM or BERT, corresponding to the aforementioned encoding methods. Furthermore, if the user question successfully matches the professional database, the matching question and answer from the professional database are input into a chatGPT-like large model, which optimizes them and outputs an answer that is easier for the user to understand. If no matching question or answer is found in the professional database, meaning the user question fails to match the professional database, the user question is directly input into the chatGPT-like large model, which generates the corresponding answer and outputs the answer generated by the large model, ensuring that the user question receives feedback. It should be noted that the output settings of large models should be adjusted to streaming output to ensure the real-time feedback results.

[0190] In some embodiments of the present application, optionally, as shown in FIG3 , the method for processing the session information may specifically include the following steps 302 to 316:

[0191] Step 302, receiving a user question;

[0192] Step 304: Translate the user's question into the professional library language through the large model and return the user's language type;

[0193] Step 306, encoding the user question;

[0194] Step 308, matching the user question with the professional database;

[0195] Step 310, determine whether the matching result is: match successful, if so, execute step 312, if not, execute step 314;

[0196] Step 312: Send the professional database matching questions and answers together into the big model to optimize the answers;

[0197] Step 314: Send the user question to the big model to directly generate the answer;

[0198] Step 316: Output the answer according to the user's language.

[0199] Among them, steps 304, 312 and the large model in step 314 are equivalent to the above-mentioned first model, steps 304, 308 and the professional library in step 312 are equivalent to the above-mentioned first database, and the professional library matching questions and answers in step 312 are equivalent to the above-mentioned second conversation information and first reply information.

[0200] In this embodiment, multi-language question-and-answer dialogues are realized with the help of the translation capabilities of the large model. Specifically, after the user question is input, the electronic device inputs the user question into the chatGPT-like large model, and translates the user question into the professional library language type through the chatGPT-like large model, and at the same time returns the user language type. On this basis, the electronic device then performs a series of operations such as encoding and similarity matching on the translated user question to obtain the corresponding answer. Furthermore, when outputting the answer to the question, the answer of the same language type should be output according to the language type of the user question to ensure the consistency of the language type of the input and output information of the question and answer process.

[0201] In one embodiment of the present application, a session information processing device is also proposed. As shown in Figure 4, Figure 4 shows a block diagram of the session information processing device 400 of the embodiment of the present application. Specifically, the session information processing device 400 may include the following receiving unit 402, processing unit 404, and display unit 406:

[0202] The receiving unit 402 is configured to receive a first input from a user, where the first input includes first session information.

[0203] Processing unit 404 is configured to match the first session information in the first database in response to the first input; if the match is successful, input the first reply information and the second session information corresponding to the first session information into the first model, so that the first model outputs the second reply information based on the first reply information and the second session information;

[0204] Display unit 406, configured to display the second reply information;

[0205] Among them, the above-mentioned first model is an intelligent dialogue model, and the above-mentioned first database is a professional knowledge base of the professional field to which the above-mentioned first conversation information belongs, such as the cooking field.

[0206] The conversation information processing device 400 provided in the embodiment of the present application is used to answer conversation information raised by users in professional fields such as cooking, and to feedback accurate and easy-to-understand reply information to the user, which can improve the correctness of intelligent question and answer results and enhance the humanized features of intelligent question and answer.

[0207] Furthermore, the session information processing device 400 provided in the present application can be specifically applied to an electronic device having data analysis and data processing functions.

[0208] In actual application, the above-mentioned electronic devices can specifically be smart devices such as smart phones, smart watches, tablet computers and laptops, and are not specifically limited here.

[0209] Specifically, the session information processing device 400 provided in the present application includes a receiving unit 402 and a processing unit 404. When a user asks questions about a professional field, such as cooking, through an electronic device, the user can directly input the first session information through a first input to the electronic device. The receiving unit 402 receives the user's first input, and the processing unit 404 obtains the first session information input by the user in response to the first input, and compares and matches the first session information with the professional knowledge base of the professional field to which it belongs, that is, the first database. The above-mentioned first database stores multiple groups of session information and reply information of the professional field to which the above-mentioned first session information belongs. On this basis, when the above-mentioned first database successfully matches the first session information input by the user, the processing unit 404 obtains the first reply information and the second session information corresponding to the above-mentioned first session information in the first database, and the first reply information is more professional. Furthermore, processing unit 404 inputs the acquired first reply information and second conversation information into the intelligent dialogue model, i.e., the first model, to optimize the first reply information, which is more professional, through the first model. Specifically, the first reply information and the second conversation information are integrated through the first model, and the first reply information is polished and corrected based on the integrated information, thereby obtaining a second reply information that is less professional and easier for the user to understand. Display unit 406 displays the second reply information to the user. In this way, by combining the professional knowledge base of the professional field and the intelligent dialogue model, intelligent answers are given to the questions raised by the user, which can ensure the professionalism of the question and answer results, thereby ensuring the correctness of the question and answer results, while being easier for the user to understand, enhancing the humanization of the question and answer results, and ensuring the question and answer effect.

[0210] Among them, the above-mentioned first input can specifically be the user's touch input to the question and answer interface in the electronic device. The touch input can specifically be single-click input, double-click input, sliding input, input along a preset trajectory, etc. Technical personnel in this field can set the specific form of the above-mentioned first input according to actual conditions, and no specific restrictions are made here.

[0211] In actual application, an information input control may be displayed in the above-mentioned question-and-answer interface. The above-mentioned first input may specifically be the user's touch input to the information input control. The processing unit 404 responds to the touch input and outputs the corresponding second reply information for the first conversation information proposed by the user.

[0212] The information input control may be a text input control or a voice input control. Furthermore, the first conversation information may be a text message or a voice message, which is not specifically limited here.

[0213] On this basis, when processing unit 404 answers the first conversation message input by the user and outputs the second reply message, if the first conversation message is a text message, processing unit 404 directly answers the first conversation message input by the user based on the text message. If the first conversation message is a voice message, processing unit 404 converts the voice message into text message and then answers the first conversation message input by the user based on the text message.

[0214] Furthermore, the first database stores multiple sets of conversation information and response information related to the professional field to which the first conversation information belongs. During the process of comparing and matching the first database with the first conversation information input by the user, processing unit 404 specifically compares the semantic similarity between the first conversation information input by the user and the multiple conversation information stored in the first database. Based on this comparison, if there is conversation information in the first database that is semantically identical or highly similar to the first conversation information, processing unit 404 determines that the first database successfully matches the first conversation information. If there is no conversation information in the first database that is semantically identical or highly similar to the first conversation information, processing unit 404 determines that the first database successfully matches the first conversation information. If there is no conversation information in the first database that is semantically identical or highly similar to the first conversation information, processing unit 404 determines that the first database failed to match the first conversation information.

[0215] In addition, in actual application, the first database may be a database pre-stored in a storage area of ​​the electronic device, or a shared database of a third-party software or website, which is not specifically limited here.

[0216] In some embodiments of the present application, optionally, the processing unit 404 is further used to: in the event of a matching failure, input the first session information into the first model, so that the first model can output third reply information based on the first session information; and the display unit 406 is further used to: display the third reply information.

[0217] In some embodiments of the present application, the first data optionally includes multiple session information. Based on this, the processing unit 404 is specifically configured to: encode the first session information; perform similarity matching on the first database and the encoded first session information; if session information with the same semantics as the first session information exists in the first database, determine that the first database and the first session information are successfully matched; if session information with the same semantics as the first session information does not exist in the first database, determine that the first database and the first session information are unmatched.

[0218] In some embodiments of the present application, optionally, the processing unit 404 is further configured to: preprocess the first session information to obtain preprocessed first session information; wherein the preprocessing includes at least one of the following: filtering out noise information in the first session information, and extracting useful features from the first session information.

[0219] In some embodiments of the present application, optionally, the processing unit 404 is further configured to: input the first session information into the first model when the first database and the first session information are in different languages; convert the language of the first session information using the first model so that the language of the first database and the first session information are the same; and the display unit 406 is specifically configured to: display the second reply information based on the language of the first session information.

[0220] In some embodiments of the present application, optionally, the above encoding processing includes at least one of the following: cosine encoding processing, vector encoding processing.

[0221] In some embodiments of the present application, optionally, the above-mentioned similarity matching algorithm includes at least one of the following: a similarity matching algorithm based on a deep learning model, a similarity matching algorithm based on cosine distance, and a similarity matching algorithm based on Euclidean distance.

[0222] In one embodiment of the present application, a session information processing system is also provided. As shown in FIG5 , FIG5 shows a block diagram of a session information processing system 500 provided in an embodiment of the present application. The session information processing system 500 includes a first model 502 , a first database 504 , and a processor 506 .

[0223] The first model 502 is an intelligent dialogue model, and the first database 504 is a professional knowledge base. The first database 504 stores multiple groups of conversation information and response information in professional fields such as cooking.

[0224] Furthermore, the processor 506 is configured to implement the steps of the method for processing session information in any of the above-mentioned embodiments. Therefore, the session information processing system 500 proposed in the third aspect of the present application has all the beneficial effects of the method for processing session information in any of the above-mentioned embodiments of the first aspect, and will not be further described here.

[0225] In one embodiment of the present application, as shown in FIG6 , a sentence generation method is also proposed, including:

[0226] Step 602: upon receiving a first question statement, identifying intent information in the first question statement;

[0227] Step 604: determining a target question statement that matches the first question statement based on the intent information;

[0228] Step 606: Generate an answer sentence based on a target corpus among the at least two corpora, wherein the target corpus matches the sentence category of the target question sentence.

[0229] In an embodiment of the present application, a sentence generation method is proposed, which is applied to a question-answering system, and the question-answering system can answer the first question sentence input by the user into the question-answering system through the sentence generation method. The question-answering system can determine the target question sentence that matches the first question sentence by means of intent recognition, and find the corresponding target corpus based on the target question sentence, and generate an answer sentence through the target corpus in the target corpus. The present application realizes the distribution of different types of questions by first performing intent recognition and classification on the first question sentence, that is, generating corresponding answer sentences through the target corpus in different target corpora, and can answer different types of question sentences through only one intent recognition, thereby reducing the waste of computing power and improving the efficiency of generating answer sentences.

[0230] In this embodiment, the question-answering system includes an intent matching module, which can identify the intent information of the first question statement and, based on the intent information, can find a target question statement corresponding to the first question statement. The target question statement found is a preset question statement, and the semantics of the target question statement are similar to or identical to the semantics of the first question statement. That is, through the identified intent information, the first question statement input by the user can be converted into a standardized target question statement.

[0231] For example, the first question statement is "How long does it take to bake a chiffon cake" or "How long does it take to make a chiffon cake?", and the recognized semantic information can be used to determine that the target question statement corresponding to the first question statement is "How long does it take to bake a chiffon cake?"

[0232] For example, the first question statement is "Why did the chiffon cake collapse?" or "What should I do if the surface of the chiffon cake is concave?". Through the recognized semantic information, it can be determined that the target question statement corresponding to the first question statement is "Why does the chiffon cake collapse after baking?"

[0233] In this embodiment, the question-answering system can find a matching target corpus in different corpora based on the sentence category. Sentence categories include but are not limited to structured question sentences and preset question sentences. The question-answering system determines the target question sentence and the sentence category corresponding to the target question sentence through intent recognition, so that the corresponding target corpus can be accurately selected. This eliminates the need to set up recognition modules that adapt to different sentence categories in the question-answering system, reduces the computing power loss of the question-answering system, and improves processing efficiency.

[0234] Structured question statements are typically simple, real-time questions, stored in a structured manner. Examples include recipe ingredients and quantities, cooking time, cooking temperature, and cooking steps. Preset question statements are typically complex questions, typically stored in a question-and-answer format. This means that the corresponding standard answer is found through the preset question statement and output as the answer statement.

[0235] In an embodiment of the present application, after receiving a first question statement, the question-answering system can determine the standardized target question statement corresponding to the first question statement through intent recognition, and based on the sentence category of the target question statement, find the corresponding target corpus, and generate an answer statement through the target corpus in the target corpus. There is no need to repeatedly process the first question statement according to different sentence categories, which reduces the energy consumption of the question-answering system and improves the efficiency of the question-answering system in outputting answer statements.

[0236] In some embodiments of the present application, optionally, as shown in FIG7 , determining a target question sentence that matches the first question sentence based on the intent information includes:

[0237] Step 702: Obtain at least two second question statements based on the intention information, where the at least two second question statements are associated with the intention information;

[0238] Step 704: Filter the target question sentence among the at least two second question sentences.

[0239] In this embodiment, it is proposed that intention information obtained through intention recognition can obtain multiple second question statements with similar semantics to the first question statement, and then the target question statement matching the first question statement is obtained by screening the multiple second question statements, thereby improving the accuracy of the screened target question statement.

[0240] In an embodiment of the present application, an intention matching module is configured in the question-answering system, which can initially recall candidate similar second question statements. After obtaining multiple second question statements, the closest target question statement can be obtained by comparing the multiple second question statements with the first question statement respectively.

[0241] Exemplarily, the intent matching module in the question-answering system uses an ES (Elastic Search)-based method and the built-in BM25 (Best Matching 25) text similarity calculation method of the question-answering system to obtain multiple second question statements that are semantically closest to the first question statement, form a candidate question set, and screen the target question statement from the candidate question set.

[0242] In an embodiment of the present application, the question-answering system can perform two-level screening through intent information, that is, first find multiple semantically similar second question statements, and then screen the target question statements that match the semantics of the first question statement among the multiple second question statements, thereby improving the accuracy of obtaining the matching target question statements.

[0243] In some embodiments of the present application, optionally, screening the target question statement from the at least two second question statements includes: obtaining semantic similarity between the first question statement and each of the at least two second question statements; and screening the target question statement from the at least two second question statements based on the semantic similarity.

[0244] This embodiment proposes a process of screening a target question sentence from a plurality of second question sentences by using semantic similarity.

[0245] In an embodiment of the present application, the semantic similarity between the first question statement and each second question statement is calculated respectively, the multiple second question statements are sorted according to the semantic similarity, and the second question statement with the highest semantic similarity is used as the target question statement, thereby improving the matching degree between the screened target question statement and the first question statement.

[0246] Among them, after obtaining multiple second question statements, the question answering system calculates the semantic similarity between the multiple second question statements and the first question statement, and then selects the second question statement with the highest semantic similarity value as the matched target question statement.

[0247] In an embodiment of the present application, after the question-answering system is able to find multiple second question statements through intent information, the question-answering system can accurately calculate multiple semantic similarities between the multiple second question statements and the first question statement, and based on the semantic similarities, find the target question statement among the multiple second question statements, further improving the matching degree between the target question statement and the first question statement.

[0248] In some embodiments of the present application, optionally, obtaining the semantic similarity between the first question statement and each of the at least two second question statements includes: obtaining a first semantic vector of the first question statement and a second semantic vector of each second question statement; and calculating the semantic similarity using the first semantic vector and the second semantic vector.

[0249] This embodiment proposes a solution of converting multiple second question statements and the first question statement into semantic vectors, and calculating the semantic similarity between each second question statement and the first question statement based on the semantic vectors.

[0250] In an embodiment of the present application, by respectively obtaining the first semantic vector and the second semantic vector corresponding to the first question statement and the second question statement, the number of the second semantic vectors is the same as the number of the second question statements, and the second semantic vectors have a one-to-one correspondence with the second question statements. By calculating the similarity between the first semantic vector and the second semantic vector, multiple semantic similarities can be obtained, and the semantic similarities have a one-to-one correspondence with the second semantic vectors.

[0251] Exemplarily, all the acquired second question statements and first question statements are converted into high-dimensional semantic vectors in the semantic space, the first question statement corresponds to the first semantic vector, and the second question semantics corresponds to the second semantic vector. The semantic similarity between the first semantic vector and each second semantic vector is calculated, and the second question statement corresponding to the second semantic vector with the highest semantic similarity is used as the target question statement.

[0252] In an embodiment of the present application, by converting both the first question statement and the second question statement into high-dimensional semantic vectors and screening the corresponding target question statement by calculating the similarity of the high-dimensional semantic vectors, the accuracy of determining the target question statement can be improved.

[0253] In some embodiments of the present application, optionally, obtaining the semantic similarity between the first question statement and each of the at least two second question statements includes: inputting the first question statement and each second question statement into a pre-trained model to determine the semantic similarity.

[0254] This embodiment proposes a process of inputting all the first question sentences and the second question sentences into a pre-training model, and calculating multiple semantic similarities between multiple second question sentences and the first question area through the training model.

[0255] Exemplarily, after obtaining multiple second question statements, the question-answering system uses a large-scale pre-trained model fine-tuned on a database of synonymous sentence pairs to calculate the semantic similarity between each of the multiple second question statements and the first question statement. For example, the multiple second question statements include: Question 1, Question 2, Question 3, and Question N, and the multiple semantic similarities corresponding to the multiple second question statements are sim1, sim2, sim3, and simN, etc. The second question statement with the highest semantic similarity value is then selected as the matched target question statement.

[0256] In an embodiment of the present application, a pre-trained model is set in the question-answering system, and when multiple second question statements are obtained, the multiple second question statements and the first question statement are input into the pre-trained model. The pre-trained model can accurately calculate multiple semantic similarities between the multiple second question statements and the first question statement.

[0257] In some embodiments of the present application, optionally, before generating an answer sentence based on a target corpus from at least two corpora, the process includes: selecting a target corpus from at least two corpora based on identification information in a target question sentence, the identification information corresponding to a sentence category; wherein the sentence category includes at least one of the following: a preset question category, a structured question category.

[0258] In this embodiment, it is proposed that the question-answering system can filter out the target corpus from at least two corpora based on the identification information corresponding to the target question sentence, that is, it is ensured that the question-answering system can determine the target corpus that matches the target question sentence corresponding to the first question sentence after determining it.

[0259] In an embodiment of the present application, the target question sentence is a question sentence preset in the question-answering system, and each target question sentence is provided with corresponding identification information. After the question-answering system determines the target question sentence corresponding to the first question sentence, it can determine the sentence category of the target question sentence by reading the identification information of the target question sentence, thereby finding the corresponding target corpus.

[0260] In an embodiment of the present application, the number of corpora configured in the question-answering system matches the number of sentence categories and the number of identification information, and the corpora, sentence categories and identification information correspond one to one.

[0261] For example, two corpora are configured in the question-answering system, so the number of sentence categories is two, and the number of identification information is also two.

[0262] In the embodiments of the present application, the sentence category includes at least one of a preset question category and a structured question category. Structured question category sentences are usually simple real-time questions, which are stored in a structured manner. Preset question category sentences are usually complex questions, which are usually stored in a question-answer correspondence format, that is, the corresponding standard answer is found through the preset question category sentence, and the standard answer is output as the answer sentence.

[0263] For example, a structured question statement may be “What is the cooking temperature of chiffon cake?”, and a preset question statement may be “Why did the chiffon cake collapse?”.

[0264] In an embodiment of the present application, by configuring corresponding identification information in the target question sentence pre-stored in the question-answering system, it is ensured that after the question-answering system obtains the target question-answer sentence that matches the first question-answer sentence, it can quickly determine the sentence category corresponding to the target question-answer sentence, and determine the target corpus among multiple corpora accordingly.

[0265] In some embodiments of the present application, optionally, as shown in FIG8 , the sentence category is a structured question category; generating an answer sentence based on a target corpus in at least two corpora includes:

[0266] Step 802, determining a target query statement based on the target question statement;

[0267] Step 804: Search the target corpus for the target corpus using the target query statement.

[0268] Step 806: Generate a response sentence based on the target corpus.

[0269] This embodiment proposes a specific process for the question-answering system to generate an answer statement when the target question statement is a structured question-type statement.

[0270] In an embodiment of the present application, when the sentence category is a structured question, a target query sentence matching the target question sentence is searched. The target query sentence is used to search for the corresponding target corpus in the target corpus corresponding to the structured question class. After the corresponding target corpus is found, the corresponding answer sentence is generated through the target corpus.

[0271] The target query statement is a structured query code.

[0272] For example, the target query statement may be “MATCH(a1:Recipe{recipe_name:“Chiffon Cake”})-[r:contain]->(a2:Ingredient{ingredient_name:“White Sugar”})RETURN contain.weight”. The target question statement corresponding to this target query statement is “the amount of white sugar in chiffon cake”.

[0273] Specifically, when the target question statement is a structured question-type statement, the template of the corresponding target query statement can be found through the target question statement. By extracting the entity nouns in the target question statement and inserting the entity nouns into the template, the target query statement can be obtained. Through the target query statement, the target corpus can be queried in the target corpus, and the answer statement can be output based on the target corpus.

[0274] For example, when the target question statement is "the amount of white sugar in chiffon cake", the target corpus found by the target query statement is "60 grams", and the answer statement generated based on the target corpus is "60 grams of white sugar is needed to make chiffon cake".

[0275] In an embodiment of the present application, when the target question statement is a structured question-type statement, the question-answering system can first determine the corresponding target query statement based on the target question statement, then find the target corpus in the corresponding target corpus through the target query statement, and then generate the corresponding answer statement based on the target corpus to ensure the degree of matching between the answer statement and the target question statement.

[0276] In some embodiments of the present application, optionally, as shown in FIG9 , determining a target query statement according to a target question statement includes:

[0277] Step 902: extract nouns and other words from the target question sentence;

[0278] Step 904: Determine a matching query template based on other words;

[0279] Step 906: Input the noun vocabulary into the query statement template to generate a target query statement.

[0280] This embodiment proposes a process of searching for a query statement template of a target query statement when the target question statement is a structured question statement, and determining the target query statement based on the query statement template.

[0281] In an embodiment of the present application, the target query statement corresponding to the structured question statement is also a structured query statement, that is, different target query statements correspond to different query statement templates. The query statement template can be determined by parsing other words in the target question statement, and then the noun words in the target question statement are inserted into the query statement template to determine the target query statement.

[0282] Among them, noun vocabulary is the entity vocabulary in the target question sentence, and other vocabulary is the non-entity vocabulary in the target sentence, that is, all vocabulary except noun vocabulary.

[0283] For example, the target question sentence is "What is the amount of white sugar in chiffon cake?" The noun vocabulary in the target question sentence is "chiffon cake" and "white sugar." Other vocabulary is "What is the amount of [ingredient name] in [recipe name]?"

[0284] In an embodiment of the present application, the question-answering system can find a query statement template that matches the target question statement by performing semantic analysis on other words.

[0285] For example, the query statement template is "MATCH(a1:Recipe{recipe_name:"

recipe name

ingredient name

[0286] In an embodiment of the present application, after determining a corresponding query statement template, the noun vocabulary in the target question statement is inserted into the query statement template to obtain a corresponding target query statement.

[0287] For example, the query statement template is "MATCH(a1:Recipe{recipe_name:"

Recipe name

Ingredient name

[0288] In an embodiment of the present application, by parsing the target question sentence of the structured question type, the noun vocabulary and other vocabulary therein can be obtained, and the corresponding query sentence template is queried through the other vocabulary. Then, the noun vocabulary is inserted into the query sentence template to obtain the corresponding target query sentence, which improves the accuracy of the target query sentence and further improves the accuracy of the target corpus found, thereby improving the degree of matching between the generated answer sentence and the first question sentence.

[0289] In some embodiments of the present application, optionally, the sentence category is a preset question category; generating an answer sentence based on a target corpus in the at least two corpora includes:

[0290] The answer sentence is searched according to the target question sentence and the preset mapping relationship, where the preset mapping relationship is the correspondence between the target question sentence and the answer sentence in the target corpus.

[0291] This embodiment proposes a specific process for the question-answering system to generate an answer statement when the target question statement is a preset question-type statement.

[0292] In an embodiment of the present application, the question-answering system stores answer statements corresponding to target question statements of a preset question class, and a preset mapping relationship between the answer statements and the target question statements of the preset question class.

[0293] The target corpus corresponding to the target question sentence of the preset question class stores a preset mapping relationship between the answer sentence and the target question sentence of the preset question class, as well as the corresponding answer sentence. After determining the target question sentence of the preset question class, the corresponding answer sentence is found using the preset mapping relationship in the target corpus and the answer sentence is output.

[0294] For example, the target question statement is "Why did the chiffon cake collapse?". Through the preset mapping relationship, the answer statements found in the target corpus include "1. It was not turned upside down to cool in time after baking, 2. The egg whites were not beaten to a hard foam state."

[0295] In an embodiment of the present application, when the target question statement is a preset question-type statement, the corresponding answer statement can be automatically found in the target corpus through the preset mapping relationship, thereby ensuring the relevance and matching degree between the answer statement and the first question statement.

[0296] In some embodiments of the present application, optionally, before identifying the intention information in the first question statement, the method further includes: upon receiving the first question statement, updating the first question statement according to vocabulary in a preset vocabulary library.

[0297] In this embodiment, it is proposed that after receiving the first question statement, the first question statement is updated by using vocabulary in a preset vocabulary library, which can improve the accuracy and search efficiency of the question-answering system in finding the target question statement through semantic information.

[0298] In an embodiment of the present application, the preset vocabulary library is a vocabulary library pre-stored in the question-answering system. Updating the vocabulary in the first question sentence by using the vocabulary in the preset vocabulary library can improve the recognition efficiency and accuracy of the semantic recognition of the first question sentence by the question-answering system.

[0299] In some embodiments of the present application, optionally, as shown in FIG10 , upon receiving the first question statement, updating the first question statement according to the vocabulary in the preset vocabulary library includes:

[0300] Step 1002: extract entity nouns in the first question sentence;

[0301] Step 1004: Based on the entity noun, search for the corresponding abstract noun in the preset vocabulary library;

[0302] Step 1006: Replace the entity nouns in the first question statement with abstract nouns to update the first question statement.

[0303] This embodiment proposes a specific process of replacing entity nouns in the first question sentence with abstract nouns in a preset vocabulary library.

[0304] Specifically, entity nouns are nouns with a smaller reference scope in the first question sentence, and abstract nouns are nouns with a larger reference scope in the preset vocabulary library. There is a semantic inclusion relationship between the abstract nouns found through entity nouns.

[0305] In the embodiment of the present application, different entity nouns may correspond to the same abstract noun in the preset vocabulary library.

[0306] For example, the entity nouns in the first question sentence are "chiffon cake", "roast chicken", and "roast lamb chops", and the abstract noun found in the preset vocabulary library is "recipe name".

[0307] For example, the entity nouns in the first question sentence are "white sugar", "chicken", and "lamb chops", and the abstract noun found in the preset vocabulary library is "ingredient name".

[0308] In an embodiment of the present application, by replacing the entity nouns in the first question statement with abstract nouns in a preset vocabulary library after receiving the first question statement, the computing power required to extract intent information from the first question statement can be reduced, and the efficiency of obtaining the intent information corresponding to the first question statement can be improved.

[0309] In one embodiment of the present application, as shown in Figure 11, a sentence generation device 1100 is also proposed, including: an identification module 1102, for identifying intent information in the first question sentence when a first question sentence is received; a determination module 1104, for determining a target question sentence that matches the first question sentence based on the intent information; a generation module 1106, for generating an answer sentence based on a target corpus of at least two corpora, wherein the target corpus matches the sentence category of the target question sentence.

[0310] In an embodiment of the present application, a sentence generation device 1100 is proposed, which is applied to a question-answering system. The question-answering system can answer the first question sentence input by the user into the question-answering system through the sentence generation device 1100. The question-answering system can determine the target question sentence that matches the first question sentence by means of intent recognition, and find the corresponding target corpus based on the target question sentence, and generate an answer sentence through the target corpus in the target corpus. The present application realizes the distribution of different types of questions by first performing intent recognition and classification on the first question sentence, that is, generating corresponding answer sentences through the target corpus in different target corpora, and can answer different types of question sentences through only one intent recognition, thereby reducing the waste of computing power and improving the efficiency of generating answer sentences.

[0311] In this embodiment, the question-answering system includes an intent matching module, which can identify the intent information of the first question statement and, based on the intent information, can find a target question statement corresponding to the first question statement. The target question statement found is a preset question statement, and the semantics of the target question statement are similar to or identical to the semantics of the first question statement. That is, through the identified intent information, the first question statement input by the user can be converted into a standardized target question statement.

[0312] In this embodiment, the question-answering system can find a matching target corpus in different corpora based on the sentence category. Sentence categories include but are not limited to structured question sentences and preset question sentences. The question-answering system determines the target question sentence and the sentence category corresponding to the target question sentence through intent recognition, so that the corresponding target corpus can be accurately selected. This eliminates the need to set up a recognition module 1102 that adapts to different sentence categories in the question-answering system, reduces the computing power loss of the question-answering system, and improves processing efficiency.

[0313] Structured question statements are typically simple, real-time questions, stored in a structured manner. Examples include recipe ingredients and quantities, cooking time, cooking temperature, and cooking steps. Preset question statements are typically complex questions, typically stored in a question-and-answer format. This means that the corresponding standard answer is found through the preset question statement and output as the answer statement.

[0314] In an embodiment of the present application, after receiving a first question statement, the question-answering system can determine the standardized target question statement corresponding to the first question statement through intent recognition, and based on the sentence category of the target question statement, find the corresponding target corpus, and generate an answer statement through the target corpus in the target corpus. There is no need to repeatedly process the first question statement according to different sentence categories, which reduces the energy consumption of the question-answering system and improves the efficiency of the question-answering system in outputting answer statements.

[0315] In some embodiments of the present application, optionally, the sentence generation device 1100 further includes: an acquisition module for acquiring at least two second question statements based on the intention information, where the at least two second question statements are associated with the intention information; and a screening module for screening the target question statement among the at least two second question statements.

[0316] In this embodiment, it is proposed that intention information obtained through intention recognition can obtain multiple second question statements with similar semantics to the first question statement, and then the target question statement matching the first question statement is obtained by screening the multiple second question statements, thereby improving the accuracy of the screened target question statement.

[0317] In an embodiment of the present application, an intention matching module is configured in the question-answering system, which can initially recall candidate similar second question statements. After obtaining multiple second question statements, the closest target question statement can be obtained by comparing the multiple second question statements with the first question statement respectively.

[0318] In an embodiment of the present application, the question-answering system can perform two-level screening through intent information, that is, first find multiple semantically similar second question statements, and then screen the target question statements that match the semantics of the first question statement among the multiple second question statements, thereby improving the accuracy of obtaining the matching target question statements.

[0319] In some embodiments of the present application, optionally, the acquisition module is specifically used to: obtain the semantic similarity between the first question statement and each second question statement in at least two second question statements; the screening module is specifically used to: screen the target question statement in at least two second question statements based on the semantic similarity.

[0320] This embodiment proposes a process of screening a target question sentence from a plurality of second question sentences by using semantic similarity.

[0321] In an embodiment of the present application, the semantic similarity between the first question statement and each second question statement is calculated respectively, the multiple second question statements are sorted according to the semantic similarity, and the second question statement with the highest semantic similarity is used as the target question statement, thereby improving the matching degree between the screened target question statement and the first question statement.

[0322] Among them, after obtaining multiple second question statements, the question answering system calculates the semantic similarity between the multiple second question statements and the first question statement, and then selects the second question statement with the highest semantic similarity value as the matched target question statement.

[0323] In an embodiment of the present application, after the question-answering system is able to find multiple second question statements through intent information, the question-answering system can accurately calculate multiple semantic similarities between the multiple second question statements and the first question statement, and based on the semantic similarities, find the target question statement among the multiple second question statements, further improving the matching degree between the target question statement and the first question statement.

[0324] In some embodiments of the present application, optionally, the acquisition module is also used to: obtain a first semantic vector of the first question statement, and a second semantic vector of each second question statement; the sentence generation device 1100 also includes: a calculation module, used to calculate semantic similarity through the first semantic vector and the second semantic vector.

[0325] This embodiment proposes a solution of converting multiple second question statements and the first question statement into semantic vectors, and calculating the semantic similarity between each second question statement and the first question statement based on the semantic vectors.

[0326] In an embodiment of the present application, by respectively obtaining the first semantic vector and the second semantic vector corresponding to the first question statement and the second question statement, the number of the second semantic vectors is the same as the number of the second question statements, and the second semantic vectors have a one-to-one correspondence with the second question statements. By calculating the similarity between the first semantic vector and the second semantic vector, multiple semantic similarities can be obtained, and the semantic similarities have a one-to-one correspondence with the second semantic vectors.

[0327] In an embodiment of the present application, by converting both the first question statement and the second question statement into high-dimensional semantic vectors and screening the corresponding target question statement by calculating the similarity of the high-dimensional semantic vectors, the accuracy of determining the target question statement can be improved.

[0328] In some embodiments of the present application, optionally, the sentence generation device 1100 further includes: an input module, configured to input the first question sentence and each second question sentence into a pre-training model to determine semantic similarity.

[0329] This embodiment proposes a process of inputting all the first question sentences and the second question sentences into a pre-training model, and calculating multiple semantic similarities between multiple second question sentences and the first question area through the training model.

[0330] In an embodiment of the present application, a pre-trained model is set in the question-answering system, and when multiple second question statements are obtained, the multiple second question statements and the first question statement are input into the pre-trained model. The pre-trained model can accurately calculate multiple semantic similarities between the multiple second question statements and the first question statement.

[0331] In some embodiments of the present application, optionally, the sentence generation device 1100 also includes: a selection module, used to select a target corpus from at least two corpora based on identification information in the target question sentence, where the identification information corresponds to the sentence category; wherein the sentence category includes at least one of the following: a preset question category and a structured question category.

[0332] In this embodiment, it is proposed that the question-answering system can filter out the target corpus from at least two corpora based on the identification information corresponding to the target question sentence, that is, it is ensured that the question-answering system can determine the target corpus that matches the target question sentence corresponding to the first question sentence after determining it.

[0333] In an embodiment of the present application, the target question sentence is a question sentence preset in the question-answering system, and each target question sentence is provided with corresponding identification information. After the question-answering system determines the target question sentence corresponding to the first question sentence, it can determine the sentence category of the target question sentence by reading the identification information of the target question sentence, thereby finding the corresponding target corpus.

[0334] In an embodiment of the present application, the number of corpora configured in the question-answering system matches the number of sentence categories and the number of identification information, and the corpora, sentence categories and identification information correspond one to one.

[0335] In the embodiments of the present application, the sentence category includes at least one of a preset question category and a structured question category. Structured question category sentences are usually simple real-time questions, which are stored in a structured manner. Preset question category sentences are usually complex questions, which are usually stored in a question-answer correspondence format, that is, the corresponding standard answer is found through the preset question category sentence, and the standard answer is output as the answer sentence.

[0336] In an embodiment of the present application, by configuring corresponding identification information in the target question sentence pre-stored in the question-answering system, it is ensured that after the question-answering system obtains the target question-answer sentence that matches the first question-answer sentence, it can quickly determine the sentence category corresponding to the target question-answer sentence, and determine the target corpus among multiple corpora accordingly.

[0337] In some embodiments of the present application, optionally, the sentence category is a structured question category; the determination module 1104 is specifically used to: determine the target query statement based on the target question statement; the sentence generation device 1100 also includes: a query module, used to query the target corpus in the target corpus through the target query statement; the generation module 1106 is specifically used to: generate an answer sentence based on the target corpus.

[0338] This embodiment proposes a specific process for the question-answering system to generate an answer statement when the target question statement is a structured question-type statement.

[0339] In an embodiment of the present application, when the sentence category is a structured question, a target query sentence matching the target question sentence is searched. The target query sentence is used to search for the corresponding target corpus in the target corpus corresponding to the structured question class. After the corresponding target corpus is found, the corresponding answer sentence is generated through the target corpus.

[0340] The target query statement is a structured query code.

[0341] In an embodiment of the present application, when the target question statement is a structured question-type statement, the question-answering system can first determine the corresponding target query statement based on the target question statement, then find the target corpus in the corresponding target corpus through the target query statement, and then generate the corresponding answer statement based on the target corpus to ensure the degree of matching between the answer statement and the target question statement.

[0342] In some embodiments of the present application, optionally, the sentence generation device 1100 also includes: an extraction module for extracting noun vocabulary and other vocabulary in the target question sentence; a determination module 1104 for determining a matching query statement template based on other vocabulary; and an input module for inputting noun vocabulary into the query statement template to generate a target query statement.

[0343] This embodiment proposes a process of searching for a query statement template of a target query statement when the target question statement is a structured question statement, and determining the target query statement based on the query statement template.

[0344] In an embodiment of the present application, the target query statement corresponding to the structured question statement is also a structured query statement, that is, different target query statements correspond to different query statement templates. The query statement template can be determined by parsing other words in the target question statement, and then the noun words in the target question statement are inserted into the query statement template to determine the target query statement.

[0345] Among them, noun vocabulary is the entity vocabulary in the target question sentence, and other vocabulary is the non-entity vocabulary in the target sentence, that is, all vocabulary except noun vocabulary.

[0346] In an embodiment of the present application, the question-answering system can find a query statement template that matches the target question statement by performing semantic analysis on other words.

[0347] In an embodiment of the present application, after determining a corresponding query statement template, the noun vocabulary in the target question statement is inserted into the query statement template to obtain a corresponding target query statement.

[0348] In an embodiment of the present application, by parsing the target question sentence of the structured question type, the noun vocabulary and other vocabulary therein can be obtained, and the corresponding query sentence template is queried through the other vocabulary. Then, the noun vocabulary is inserted into the query sentence template to obtain the corresponding target query sentence, which improves the accuracy of the target query sentence and further improves the accuracy of the target corpus found, thereby improving the degree of matching between the generated answer sentence and the first question sentence.

[0349] In some embodiments of the present application, optionally, the sentence category is a preset question category; the sentence generation device 1100 also includes: a search module, used to search for answer sentences based on the target question sentence and the preset mapping relationship, and the preset mapping relationship is the correspondence between the target question sentence and the answer sentence in the target corpus.

[0350] This embodiment proposes a specific process for the question-answering system to generate an answer statement when the target question statement is a preset question-type statement.

[0351] In an embodiment of the present application, the question-answering system stores answer statements corresponding to target question statements of a preset question class, and a preset mapping relationship between the answer statements and the target question statements of the preset question class.

[0352] The target corpus corresponding to the target question sentence of the preset question class stores a preset mapping relationship between the answer sentence and the target question sentence of the preset question class, as well as the corresponding answer sentence. After determining the target question sentence of the preset question class, the corresponding answer sentence is found using the preset mapping relationship in the target corpus and the answer sentence is output.

[0353] In an embodiment of the present application, when the target question statement is a preset question-type statement, the corresponding answer statement can be automatically found in the target corpus through the preset mapping relationship, thereby ensuring the relevance and matching degree between the answer statement and the first question statement.

[0354] In some embodiments of the present application, optionally, the sentence generation device 1100 further includes: an updating module, configured to update the first question sentence according to vocabulary in a preset vocabulary library when the first question sentence is received.

[0355] In this embodiment, it is proposed that after receiving the first question statement, the first question statement is updated by using vocabulary in a preset vocabulary library, which can improve the accuracy and search efficiency of the question-answering system in finding the target question statement through semantic information.

[0356] In an embodiment of the present application, the preset vocabulary library is a vocabulary library pre-stored in the question-answering system. Updating the vocabulary in the first question sentence by using the vocabulary in the preset vocabulary library can improve the recognition efficiency and accuracy of the semantic recognition of the first question sentence by the question-answering system.

[0357] In some embodiments of the present application, optionally, an extraction module is used to extract entity nouns in the first question statement; a search module is used to search for corresponding abstract nouns in a preset vocabulary library based on the entity nouns; and an update module is used to replace the entity nouns in the first question statement with abstract nouns to update the first question statement.

[0358] This embodiment proposes a specific process of replacing entity nouns in the first question sentence with abstract nouns in a preset vocabulary library.

[0359] Specifically, entity nouns are nouns with a smaller reference scope in the first question sentence, and abstract nouns are nouns with a larger reference scope in the preset vocabulary library. There is a semantic inclusion relationship between the abstract nouns found through entity nouns.

[0360] In the embodiment of the present application, different entity nouns may correspond to the same abstract noun in the preset vocabulary library.

[0361] In an embodiment of the present application, by replacing the entity nouns in the first question statement with abstract nouns in a preset vocabulary library after receiving the first question statement, the computing power required to extract intent information from the first question statement can be reduced, and the efficiency of obtaining the intent information corresponding to the first question statement can be improved.

[0362] In one embodiment of the present application, as shown in Figure 12, a sentence generation device 1200 is also proposed, including: a processor 1202 and a memory 1204, in which a program or instruction is stored; the processor 1202 executes the program or instruction stored in the memory 1204 to implement the steps of the sentence generation method in any of the above embodiments, and thus has all the beneficial technical effects of the sentence generation method in any of the above embodiments, which will not be elaborated here.

[0363] In one embodiment of the present application, an electronic device is also provided. As shown in FIG13 , FIG13 shows a block diagram of the structure of the electronic device 1300 provided in the embodiment of the present application. The electronic device 1300 includes:

[0364] Memory 1302 , where programs or instructions are stored;

[0365] The processor 1304 implements the steps of the method for processing session information in any of the above embodiments when executing the above program or instruction.

[0366] The electronic device 1300 provided in this embodiment includes a memory 1302 and a processor 1304. When the program or instructions in the memory 1302 are executed by the processor 1304, the steps of the method for processing session information in any of the aforementioned embodiments are implemented. Alternatively, when the program or instructions in the memory 1302 are executed by the processor 1304, the steps of the method for generating a statement in any of the aforementioned embodiments are implemented. Therefore, the electronic device 1300 has all the beneficial effects of the method for processing session information in any of the aforementioned embodiments, or the method for generating a statement in any of the aforementioned embodiments, and further description thereof will not be given here.

[0367] Specifically, the memory 1302 and the processor 1304 may be connected via a bus or other means. The processor 1304 may include one or more processing units, and the processor 1304 may be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other chips.

[0368] In actual application, the electronic device 1300 may be a smart phone, a smart watch, a tablet computer, a laptop computer, or other smart device with data analysis and data processing functions, and no specific limitation is imposed here.

[0369] In one embodiment of the present application, a readable storage medium is further provided. A program or instruction is stored on the medium. When executed by a processor, the program or instruction implements the steps of the method for processing session information in any of the above embodiments, or when executed by a processor, the program or instruction implements the steps of the method for generating statements in any of the above embodiments.

[0370] The readable storage medium provided in the embodiments of the present application, when executed by a processor, stores a program or instructions that implement the steps of the method for processing session information in any of the aforementioned embodiments, or the steps of the method for generating a statement in any of the aforementioned embodiments. Therefore, the readable storage medium possesses all the beneficial effects of the method for processing session information in any of the aforementioned embodiments, or the steps of the method for generating a statement in any of the aforementioned embodiments, and no further details are given here.

[0371] Specifically, the above-mentioned readable storage medium may include any medium capable of storing or transmitting information. Examples of readable storage media include electronic circuits, semiconductor memory devices, read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), flash memory, erasable ROM (EROM), magnetic tape, floppy disk, optical disc, hard disk, optical fiber medium, radio frequency (RF) link, optical data storage device, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0372] In one embodiment of the present application, as shown in Figure 14, a cooking device 1400 is also proposed, including: a sentence generating device 1200 as in any of the above embodiments, and / or a readable storage medium as in any of the above embodiments, and thus has all the beneficial technical effects of the sentence generating device 1200 in any of the above embodiments, and / or the readable storage medium in any of the above embodiments, and will not be elaborated on here.

[0373] In some embodiments of the present application, optionally, the cooking device 1400 includes any one of the following: a microwave oven, a microwave oven-steamer-oven, and an air fryer.

[0374] In this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance, unless otherwise expressly specified or limited. Terms such as "connect," "install," and "fix" should be interpreted broadly. For example, "connect" can refer to a fixed connection, a detachable connection, or an integral connection; it can refer to a direct connection or an indirect connection through an intermediary. Those skilled in the art will understand the specific meanings of these terms in this application based on specific circumstances.

[0375] Throughout this specification, terms such as "one embodiment," "some embodiments," and "specific embodiments" mean that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present application. In this specification, schematic representations 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 any one or more embodiments or examples.

[0376] In addition, the technical solutions between the various embodiments of the present application can be combined with each other, but it must be based on the fact that ordinary technicians in this field can implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0377] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for processing session information, wherein: include: receiving a first input from a user, wherein the first input includes first session information; In response to the first input, matching the first session information in a first database; In the case of successful matching, the second session information and the first reply information corresponding to the first session information are input into the first model, so that the first model outputs the second reply information according to the second session information and the first reply information; displaying the second reply information; The first database is a professional knowledge base of the professional field in which the first conversation information is located, and the first model is an intelligent dialogue model.

2. The method for processing session information according to claim 1, wherein: After matching the first session information in the first database, the method for processing the session information further includes: In the case where the first database matching fails, inputting the first session information into the first model, so that the first model outputs third reply information according to the first session information; The third reply information is displayed.

3. The method for processing session information according to claim 1, wherein: The first data includes a plurality of session information, and matching the first session information in the first database includes: performing encoding processing on the first session information; performing similarity matching on the encoded first session information and the plurality of session information in the first database; If there is session information in the first database that has the same semantics as the first session information, determining that the first session information successfully matches the first database; In the case where the semantics of any session information in the first database are different from those of the first session information, it is determined that the first session information fails to match the first database.

4. The method for processing session information according to claim 3, wherein: Before encoding the first session information, the method for processing the session information further includes: Preprocessing the first session information to obtain preprocessed first session information; The preprocessing includes at least one of the following: extracting useful features from the first session information, and filtering out noise information from the first session information.

5. The method for processing session information according to claim 3, wherein: Before encoding the first session information, the method for processing the session information further includes: In a case where the language type of the first session information does not match the language type of the first database, inputting the first session information into the first model so that the first model converts the language type of the first session information into the language type of the first database; The displaying of the second reply information includes: The second reply information is displayed according to the language type of the first conversation information.

6. The method for processing session information according to any one of claims 3 to 5, wherein: The encoding process includes at least one of the following: vector encoding process and cosine encoding process.

7. The method for processing session information according to any one of claims 3 to 5, wherein: The similarity matching algorithm includes at least one of the following: a similarity matching algorithm based on Euclidean distance, a similarity matching algorithm based on cosine distance, and a similarity matching algorithm based on a deep learning model.

8. A session information processing device, wherein: include: A receiving unit, configured to receive a first input from a user, wherein the first input includes first session information; a processing unit, configured to match the first session information in a first database in response to the first input; if the match is successful, input the second session information and the first reply information corresponding to the first session information into a first model, so that the first model outputs the second reply information according to the second session information and the first reply information; A display unit, configured to display the second reply information; The first database is a professional knowledge base of the professional field in which the first conversation information is located, and the first model is an intelligent dialogue model.

9. A session information processing system, wherein: include: a first database, wherein the first database stores multiple sets of conversation information and reply information; A first model, wherein the first model is an intelligent dialogue model; A processor, wherein the processor is used to implement the steps of the method for processing session information according to any one of claims 1 to 7.

10. A sentence generation method, wherein: include: Upon receiving a first question statement, identifying intent information in the first question statement; Determining, according to the intention information, a target question sentence that matches the first question sentence; An answer sentence is generated based on a target corpus of at least two corpora, wherein the target corpus matches a sentence category of the target question sentence.

11. The sentence generation method according to claim 10, wherein: The determining, according to the intention information, a target question sentence matching the first question sentence includes: Acquire at least two second question statements according to the intention information, wherein the at least two second question statements are associated with the intention information; The target question sentence among at least two of the second question sentences is selected.

12. The sentence generation method according to claim 11, wherein: The step of screening the target question statement from at least two of the second question statements includes: Obtaining a semantic similarity between the first question statement and each of the at least two second question statements; Based on the semantic similarity, the target question sentences in at least two of the second question sentences are screened.

13. The sentence generation method according to claim 12, wherein: The obtaining of the semantic similarity between the first question statement and each of the at least two second question statements includes: Obtaining a first semantic vector of the first question sentence and a second semantic vector of each of the second question sentences; The semantic similarity is calculated using the first semantic vector and the second semantic vector.

14. The sentence generation method according to claim 12, wherein: The obtaining of the semantic similarity between the first question statement and each of the at least two second question statements includes: The first question sentence and each of the second question sentences are input into a pre-trained model to determine the semantic similarity.

15. The sentence generation method according to any one of claims 10 to 14, wherein: Before generating the answer sentence based on the target corpus of the at least two corpora, the method includes: Selecting the target corpus from the at least two corpora based on identification information in the target question sentence, wherein the identification information corresponds to the sentence category; The sentence category includes at least one of the following: a preset question category and a structured question category.

16. The sentence generation method according to claim 15, wherein: The sentence category is the structured question category; The generating of the answer sentence based on the target corpus of the at least two corpora comprises: Determine a target query statement according to the target question statement; Using the target query statement, searching for target corpus in the target corpus; The answer sentence is generated according to the target corpus.

17. The sentence generation method according to claim 16, wherein: Determining a target query statement according to the target question statement includes: Extracting nouns and other words from the target question sentence; Determining a matching query statement template according to the other words; The noun vocabulary is input into the query sentence template to generate the target query sentence.

18. The sentence generation method according to claim 15, wherein: The sentence category is the preset question category; The generating of the answer sentence based on the target corpus of the at least two corpora comprises: The answer sentence is searched according to the target question sentence and a preset mapping relationship, where the preset mapping relationship is a corresponding relationship between the target question sentence and the answer sentence in the target corpus.

19. The sentence generation method according to any one of claims 10 to 14, wherein: Before identifying the intention information in the first question sentence, the method further includes: When the first question statement is received, the first question statement is updated according to the vocabulary in the preset vocabulary library.

20. The sentence generation method according to claim 19, wherein: When the first question statement is received, updating the first question statement according to the vocabulary in the preset vocabulary library includes: Extracting entity nouns in the first question sentence; Based on the entity noun, searching for a corresponding abstract noun in the preset vocabulary library; The entity noun in the first question statement is replaced with the abstract noun to update the first question statement.

21. A sentence generating device, wherein: include: an identification module, configured to identify intention information in a first question statement upon receiving the first question statement; A determination module, configured to determine a target question sentence matching the first question sentence according to the intention information; A generating module is used to generate an answer sentence based on a target corpus of at least two corpora, wherein the target corpus matches the sentence category of the target question sentence.

22. A sentence generating device, wherein: include: A processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method according to any one of claims 10 to 20 are implemented.

23. An electronic device, wherein: The method comprises a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of the method for processing session information as described in any one of claims 1 to 7 are implemented, or when the program or instruction is executed by the processor, the steps of the method for generating a sentence as described in any one of claims 10 to 20 are implemented.

24. A readable storage medium, wherein: The readable storage medium stores a program or an instruction, which, when executed by a processor, implements the steps of the method for processing session information according to any one of claims 1 to 7, or, when executed by a processor, implements the steps of the method for generating a sentence according to any one of claims 10 to 20.

25. A cooking device, wherein: include: The sentence generating device as claimed in claim 21 or 22; and / or The readable storage medium of claim 24.

Images