Intelligent question and answer method and computing device

Abstract

The embodiment of the invention provides an intelligent question and answer method and computing equipment. The method comprises the following steps: acquiring a multilingual text; performing text language recognition and translation processing on the multilingual text to obtain a question text corresponding to a target language; according to the target language, determining an answer text corresponding to the question text from a multi-language knowledge base; wherein the multilingual knowledge base is constructed according to the preset knowledge points and the large language model. Through the above mode, the computing device can fully understand the multilingual text, and the accuracy of determining the answer text is improved.

Description

Technical Field

[0001] This application relates to the technical field of computing devices, and particularly to an intelligent question-answering method and a computing device. Background Art

[0002] With the blessing of the era of digital and international development in both directions, intelligent question-answering is becoming more and more common. During the question-answering process, the language expression preferences of different users are different. Especially in large enterprises, users often use a combination of Chinese and English for question-answering.

[0003] In related technologies, during the intelligent question-answering process, a computing device can use monolingual language materials to perform matching processing on the obtained text (such as speech text) to obtain an answer text.

[0004] However, in the case where the text input by the user is a multi-language mixed text, the methods in related technologies may not be able to fully understand the multi-language text input by the user, resulting in a low accuracy rate of the determined answer text in the methods of related technologies. Summary of the Invention

[0005] Embodiments of this application provide an intelligent question-answering method and a computing device, which improve the accuracy rate of determining the answer text.

[0006] In a first aspect, embodiments of this application provide an intelligent question-answering method, and the method includes:

[0007] Obtain a multi-language text;

[0008] Perform text language recognition and translation processing on the multi-language text to obtain a question text corresponding to the target language;

[0009] Determine an answer text corresponding to the question text from a multi-language knowledge base according to the target language; wherein, the multi-language knowledge base is constructed according to preset knowledge points and a large language model.

[0010] In this solution, a computing device can obtain a multi-language text, perform text language recognition and translation processing on the multi-language text to obtain a question text corresponding to the target language. The computing device can determine an answer text corresponding to the question text from the multi-language knowledge base according to the target language; wherein, the multi-language knowledge base is constructed according to preset knowledge points and a large language model. By the above method of performing text language recognition and translation processing on the multi-language text, the languages can be aligned, enabling the computing device to fully understand the content of the multi-language text. Furthermore, based on the question text corresponding to the target language, the multi-language knowledge base is retrieved to obtain the answer text, improving the accuracy rate of determining the answer text.

[0011] In one implementation, text language identification and translation processing are performed on multilingual text to obtain problem text corresponding to the target language, including:

[0012] The multilingual text is tokenized through a language identification model to obtain multiple first keywords, and the language corresponding to each first keyword among the multiple first keywords is determined;

[0013] According to the language corresponding to each first keyword and the target language, the multilingual text is translated through a language translation model to obtain the problem text.

[0014] In this solution, the computing device can obtain multilingual text. The computing device can tokenize the multilingual text through a language identification model to obtain multiple first keywords, and determine the language corresponding to each first keyword among the multiple first keywords. The computing device can translate the multilingual text through a language translation model according to the language corresponding to each first keyword and the target language to obtain the problem text. The computing device can determine the answer text corresponding to the problem text from the multilingual knowledge base according to the target language; wherein, the multilingual knowledge base is constructed based on preset knowledge points and large language models. By the above method of determining the language corresponding to each first keyword among the multiple first keywords included in the multilingual text, and translating the multilingual text according to the language corresponding to each first keyword and the target language, language alignment can be performed, so that the computing device can fully understand the content of the multilingual text, and then retrieve the multilingual knowledge base based on the problem text corresponding to the target language to obtain the answer text, improving the accuracy of determining the answer text.

[0015] In one implementation, the method further includes:

[0016] According to the language corresponding to each first keyword, determine the language with the largest number of corresponding first keywords among the multiple languages as the target language; or,

[0017] According to the language corresponding to each first keyword, determine the language among the multiple languages with the number of corresponding first keywords greater than the first threshold as the target language; or,

[0018] Determine the preset language as the target language.

[0019] In this solution, the computing device can determine the language with the largest number of corresponding first keywords among the multiple languages as the target language according to the language corresponding to each first keyword. The computing device can also determine the language among the multiple languages with the number of corresponding first keywords greater than the first threshold as the target language according to the language corresponding to each first keyword. The computing device can also determine the preset language as the target language. By the above method, the speed of determining the target language is improved.

[0020] In one implementation, according to the target language, the answer text corresponding to the question text is determined from a multilingual knowledge base, including:

[0021] Perform vectorization processing on the question text to obtain a first question vector;

[0022] According to the first question vector, query the multilingual knowledge base, and determine the target question vector from multiple initial question vectors; the multilingual knowledge base includes multiple initial question vectors and the initial answer text corresponding to each initial question vector; among them, the multiple initial question vectors correspond to the target language;

[0023] Determine the target answer text corresponding to the target question vector as the answer text.

[0024] In this solution, the computing device can perform vectorization processing on the question text to obtain a first question vector. The computing device can query the multilingual knowledge base according to the first question vector and determine the target question vector from multiple initial question vectors; the multilingual knowledge base includes multiple initial question vectors and the initial answer text corresponding to each initial question vector; the multiple initial question vectors correspond to the target language. The computing device can determine the target answer text corresponding to the target question vector as the answer text. Through the above method, the accuracy of determining the answer text from the multilingual knowledge base is improved.

[0025] In one implementation, according to the first question vector, query the multilingual knowledge base, and determine the target question vector from multiple initial question vectors, including:

[0026] Determine the matching degree corresponding to each initial question vector according to the multiple initial question vectors and the first question vector;

[0027] Determine the target question vector from the multiple initial question vectors according to the matching degrees corresponding to the multiple initial question vectors.

[0028] In this solution, the computing device can determine the target question vector from multiple initial question vectors by calculating the matching degrees between the multiple initial question vectors in the multilingual knowledge base and the first question vector, improving the accuracy and efficiency of determining the target text vector.

[0029] In one implementation, according to the matching degrees corresponding to the multiple initial question vectors, determine the target question vector from the multiple initial question vectors, including:

[0030] Obtain the corresponding preset matching degree threshold according to the target language;

[0031] Determine the target question vector from the multiple initial question vectors according to the matching degrees corresponding to the multiple initial question vectors and the preset matching degree threshold.

[0032] In this solution, the computing device can determine a target problem vector from multiple initial problem vectors according to the matching degrees corresponding to the multiple initial problem vectors and a preset matching degree threshold corresponding to the target language. In the above manner, the matching degree between the selected target problem vector and the first problem vector is ensured, and thus the accuracy of the determined answer text is ensured.

[0033] In one implementation, the method further includes:

[0034] Perform word segmentation on the answer text to obtain multiple second keywords;

[0035] Perform translation processing on at least one of the second keywords in the answer text to obtain a multilingual answer text.

[0036] In this solution, the computing device can perform word segmentation on the answer text to obtain multiple second keywords, and perform translation processing on at least one of the second keywords in the answer text to obtain a multilingual answer text. By converting the answer text into a multilingual answer text in the above manner, the multilingual answer text can meet the personalized needs of users and improve the user's question-answering experience.

[0037] In one implementation, the method further includes:

[0038] Obtain multiple text sentences;

[0039] Perform word segmentation on the text sentences to obtain multiple third keywords;

[0040] Perform translation processing on at least one of the third keywords in the text sentences to obtain a sample text sentence corresponding to the text sentences;

[0041] Obtain label information corresponding to the sample text sentence; the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword;

[0042] Train an initial language recognition model according to the multiple sample text sentences and the label information corresponding to each sample text sentence to obtain a language recognition model.

[0043] In this solution, the computing device can obtain multiple text statements. The computing device can perform word segmentation on the text statements to obtain multiple third keywords, and perform translation processing on at least one of the third keywords in the text statements to obtain a sample text statement corresponding to the text statement. The computing device can obtain the label information corresponding to the sample text statement; the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword. The computing device can train an initial language recognition model based on multiple sample text statements and the label information corresponding to each sample text statement to obtain a language recognition model. Through the above model training method, the language recognition accuracy of the language recognition model can be improved.

[0044] In one implementation, the method further includes:

[0045] Obtain multiple text statements and the languages corresponding to the text statements;

[0046] Perform word segmentation on the text statements to obtain multiple third keywords;

[0047] Perform translation processing on at least one of the third keywords in the text statements to obtain a sample text statement corresponding to the text statement;

[0048] Obtain the label information corresponding to the sample text statement; the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword;

[0049] Train an initial language translation model based on multiple text statements, the languages corresponding to each text statement, sample text statements, and the label information corresponding to each sample text statement to obtain a language translation model.

[0050] In this solution, the computing device can obtain multiple text statements and the languages corresponding to the text statements. The computing device can perform word segmentation on the text statements to obtain multiple third keywords. The computing device can perform translation processing on at least one of the third keywords in the text statements to obtain a sample text statement corresponding to the text statement, and obtain the label information corresponding to the sample text statement; the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword. The computing device can train an initial language translation model based on multiple text statements, the languages corresponding to each text statement, sample text statements, and the label information corresponding to each sample text statement to obtain a language translation model. Through the above model training method, the translation accuracy of the language translation model can be improved.

[0051] In one implementation, the method further includes:

[0052] Obtain the initial library versions corresponding to multiple languages; the initial library versions include multiple initial question texts and the initial answer texts corresponding to each initial question text;

[0053] Perform vectorization processing on the initial question texts in the initial libraries corresponding to multiple languages to obtain the knowledge base versions corresponding to multiple languages; the knowledge base corresponding to a language includes multiple initial question vectors and the initial answer texts corresponding to each initial question vector;

[0054] Construct a multilingual knowledge base based on the knowledge base versions corresponding to multiple languages.

[0055] In this solution, the computing device can obtain the initial library versions corresponding to multiple languages. Among them, the initial library includes multiple initial question texts and the initial answer texts corresponding to each initial question text. The computing device can perform vectorization processing on the initial question texts in multiple initial library versions to obtain the knowledge base versions corresponding to multiple languages. The computing device can construct a multilingual knowledge base based on the knowledge base versions corresponding to multiple languages. By pre-constructing the multilingual knowledge base, in scenarios where intelligent question answering for multilingual texts is required, the answer text corresponding to the question text can be quickly determined from the multilingual knowledge base according to the target language. That is to say, by pre-constructing the multilingual knowledge base, different language question-and-answer scenarios can be better handled, and the query accuracy of the answer text is improved.

[0056] In one implementation, obtaining the initial library versions corresponding to multiple languages includes:

[0057] Obtain preset knowledge points; where the preset knowledge points correspond to the first language;

[0058] According to the preset knowledge points and the large language model, obtain the initial library version corresponding to the first language; the initial question texts and initial answer texts in the initial library version corresponding to the first language correspond to the first language;

[0059] Perform translation processing on the initial library version corresponding to the first language to obtain the initial library version corresponding to the second language; the initial question texts and initial answer texts in the initial library version corresponding to the second language correspond to the second language.

[0060] In this solution, the computing device can obtain preset knowledge points (the preset knowledge points correspond to the first language). The computing device can obtain the initial library version corresponding to the first language according to the preset knowledge points and the large language model; the initial question text and the initial answer text in the initial library version corresponding to the first language correspond to the first language. The computing device can perform translation processing on the initial library version corresponding to the first language to obtain the initial library version corresponding to the second language; the initial question text and the initial answer text in the initial library version corresponding to the second language correspond to the second language. Through the above method using the large language model, the initial library version corresponding to the first language can be obtained quickly. By the above method of performing translation processing on the initial library version corresponding to the first language to obtain the initial library version corresponding to the second language, the efficiency of obtaining the initial library version corresponding to the second language is improved, the usage rate of the large language model is reduced, and thus the resource consumption is reduced.

[0061] In a second aspect, an embodiment of the present application provides an intelligent question answering device, which includes:

[0062] An acquisition module, configured to acquire multi-language texts;

[0063] A processing module, configured to perform text language identification and translation processing on the multi-language texts to obtain a question text corresponding to the target language;

[0064] The processing module is further configured to determine an answer text corresponding to the question text from a multi-language knowledge base according to the target language; wherein, the multi-language knowledge base is constructed according to preset knowledge points and a large language model.

[0065] The intelligent question answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0066] In one implementation, the processing module is specifically configured to:

[0067] Perform word segmentation processing on the multi-language texts through a language identification model to obtain a plurality of first keywords, and determine the language corresponding to each first keyword in the plurality of first keywords;

[0068] Perform translation processing on the multi-language texts through a language translation model according to the language corresponding to each first keyword and the target language to obtain the question text.

[0069] The intelligent question answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0070] In one implementation, the processing module is further configured to:

[0071] According to the languages corresponding to the first keywords, determine the language with the largest number of first keywords corresponding to it among multiple languages as the target language; or,

[0072] According to the languages corresponding to the first keywords, determine the languages among multiple languages whose number of first keywords corresponding to them is greater than the first threshold as the target languages; or,

[0073] Determine the preset language as the target language.

[0074] The intelligent question-and-answer device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0075] In one implementation, the processing module is specifically configured to:

[0076] Perform vectorization processing on the question text to obtain a first question vector;

[0077] According to the first question vector, query the multilingual knowledge base, and determine a target question vector from multiple initial question vectors; the multilingual knowledge base includes multiple initial question vectors and the initial answer texts corresponding to the initial question vectors; among them, the multiple initial question vectors correspond to the target language;

[0078] Determine the target answer text corresponding to the target question vector as the answer text.

[0079] The intelligent question-and-answer device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0080] In one implementation, the processing module is specifically configured to:

[0081] Determine the matching degree corresponding to each initial question vector according to the multiple initial question vectors and the first question vector;

[0082] Determine the target question vector from the multiple initial question vectors according to the matching degrees corresponding to the multiple initial question vectors.

[0083] The intelligent question-and-answer device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0084] In one implementation, the processing module is specifically configured to:

[0085] Obtain the corresponding preset matching degree threshold according to the target language;

[0086] Determine the target question vector from the multiple initial question vectors according to the matching degrees corresponding to the multiple initial question vectors and the preset matching degree threshold.

[0087] The intelligent question-answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0088] In one implementation, the processing module is further configured to:

[0089] Perform word segmentation on the answer text to obtain a plurality of second keywords;

[0090] Perform translation processing on at least one second keyword in the answer text to obtain a multi-language answer text.

[0091] The intelligent question-answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0092] In one implementation, the processing module is further configured to:

[0093] Obtain a plurality of text statements;

[0094] Perform word segmentation on the text statements to obtain a plurality of third keywords;

[0095] Perform translation processing on at least one third keyword in the text statements to obtain a sample text statement corresponding to the text statements;

[0096] Obtain label information corresponding to the sample text statements; the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword;

[0097] Train an initial language recognition model according to the plurality of sample text statements and the label information corresponding to each sample text statement to obtain a language recognition model.

[0098] The intelligent question-answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0099] In one implementation, the processing module is further configured to:

[0100] Obtain a plurality of text statements and the languages corresponding to the text statements;

[0101] Perform word segmentation on the text statements to obtain a plurality of third keywords;

[0102] Perform translation processing on at least one third keyword in the text statements to obtain a sample text statement corresponding to the text statements;

[0103] Obtain the tag information corresponding to the sample text sentence; the tag information includes the language types corresponding to at least one translated third keyword and the language types corresponding to at least one untranslated third keyword;

[0104] Train the initial language translation model based on multiple text sentences, the language types corresponding to each text sentence and the sample text sentence, and the tag information corresponding to each sample text sentence, to obtain the language translation model.

[0105] The intelligent question answering device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described here again.

[0106] In one implementation, the processing module is further configured to:

[0107] Obtain the initial library versions corresponding to multiple language types; the initial library versions include multiple initial question texts and the initial answer texts corresponding to each initial question text;

[0108] Perform vectorization processing on the initial question texts in the initial libraries corresponding to multiple language types to obtain the knowledge base versions corresponding to multiple language types; the knowledge base corresponding to a language type includes multiple initial question vectors and the initial answer texts corresponding to each initial question vector;

[0109] Construct a multilingual knowledge base according to the knowledge base versions corresponding to multiple language types.

[0110] The intelligent question answering device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described here again.

[0111] In one implementation, the processing module is specifically configured to:

[0112] Obtain a preset knowledge point; wherein, the preset knowledge point corresponds to the first language type;

[0113] Obtain the initial library version corresponding to the first language type according to the preset knowledge point and the large language model; the initial question texts and the initial answer texts in the initial library version corresponding to the first language type correspond to the first language type;

[0114] Perform translation processing on the initial library version corresponding to the first language type to obtain the initial library version corresponding to the second language type; the initial question texts and the initial answer texts in the initial library version corresponding to the second language type correspond to the second language type.

[0115] The intelligent question answering device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described here again.

[0116] In a third aspect, an embodiment of the present application provides a computing device, which includes a memory and a processor;

[0117] The memory is coupled to the processor;

[0118] The memory is used to store computer instructions;

[0119] The processor is used to execute the computer instructions so that the computing device implements the method of the first aspect.

[0120] The computing device provided by the embodiment of the present application can execute the technical solutions shown in the above method embodiments, and its implementation principle and beneficial effects are similar, which will not be elaborated here.

[0121] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the method of the first aspect.

[0122] When the computer-executable instructions in the computer-readable storage medium provided by the embodiment of the present application are executed by a processor, the technical solutions shown in the above method embodiments can be implemented, and its implementation principle and beneficial effects are similar, which will not be elaborated here.

[0123] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it is used to implement the method of the first aspect.

[0124] When the computer program in the computer program product provided by the embodiment of the present application is executed by a processor, the technical solutions shown in the above method embodiments can be implemented, and its implementation principle and beneficial effects are similar, which will not be elaborated here. Description of the Drawings

[0125] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0126] Figure 1 It is a schematic diagram of the scenario of an intelligent question and answer method provided by an embodiment of the present application;

[0127] Figure 2 It is a schematic flowchart of the first embodiment of an intelligent question and answer method provided by an embodiment of the present application;

[0128] Figure 3a It is a schematic flowchart of the second embodiment of an intelligent question and answer method provided by an embodiment of the present application;

[0129] Figure 3b A schematic diagram of a scenario for determining the language of each first keyword provided by an embodiment of the present application;

[0130] Figure 3c A schematic diagram of a process for training an initial language recognition model provided by an embodiment of the present application;

[0131] Figure 3d A schematic diagram of a scenario for model training - model testing provided by an embodiment of the present application;

[0132] Figure 3e A schematic diagram of a scenario for obtaining a problem text provided by an embodiment of the present application;

[0133] Figure 3f A schematic diagram of a process for training a language translation model provided by an embodiment of the present application;

[0134] Figure 3g Another schematic diagram of a scenario for model training - model testing provided by an embodiment of the present application;

[0135] Figure 4 A schematic diagram of a process for the third embodiment of an intelligent question - answering method provided by an embodiment of the present application;

[0136] Figure 5 A schematic diagram of a process for the fourth embodiment of an intelligent question - answering method provided by an embodiment of the present application;

[0137] Figure 6 A schematic diagram of the structure of an intelligent question - answering device provided by an embodiment of the present application;

[0138] Figure 7 A schematic diagram of the structure of a computing device provided by an embodiment of the present application. Detailed implementation manners

[0139] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application under the inspiration of this embodiment belong to the scope of protection of the present application.

[0140] In the description and claims of this application and the above-mentioned drawings, the terms "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0141] An embodiment of this application provides an intelligent question-and-answer method. A computing device can obtain multilingual text, perform text language recognition and translation processing on the multilingual text, and obtain a question text corresponding to the target language. The computing device can determine an answer text corresponding to the question text from a multilingual knowledge base according to the target language; where the multilingual knowledge base is constructed based on preset knowledge points and a large language model.

[0142] Through the above method, the computing device can fully understand the multilingual text, improving the accuracy of determining the answer text.

[0143] The intelligent question-and-answer method of the embodiment of this application will be described in detail below.

[0144] Figure 1 This is a schematic diagram of the scenario of an intelligent question-and-answer method provided by an embodiment of this application. As Figure 1 shown, this scenario includes a terminal device 10 and a computing device 20.

[0145] It should be noted that the computing device 20 can be a server or a server cluster.

[0146] When the computing device 20 is a server, from an architectural perspective, the server can be a rack server, a high-density server, a tower server, or a whole cabinet server; from a functional perspective, the server can be a general-purpose server or an artificial intelligence server (AI (artificial intelligence) server), etc. Exemplarily, the artificial intelligence server can be an image processing server (GPU (graphics processing unit) server).

[0147] It should also be noted that the terminal device 20 can be, but is not limited to, various personal computers, laptop computers, smartphones, and tablet computers.

[0148] In Figure 1In the scenario shown, the computing device 20 can obtain multilingual text. In one implementation, the multilingual text can be sent by the terminal device 10.

[0149] The computing device 20 can perform text language identification and translation processing on the multilingual text to obtain problem text corresponding to the target language.

[0150] The computing device can determine answer text corresponding to the problem text from the multilingual knowledge base according to the target language. The multilingual knowledge base is constructed based on preset knowledge points and a large speech model. In one implementation, the terminal device 10 can obtain the answer text sent by the computing device 20.

[0151] It should be noted that Figure 1 is a schematic diagram of the scenario provided by the embodiments of the present application. The embodiments of the present application do not limit Figure 1 the actual forms of various devices included therein, nor do they limit Figure 1 the interaction methods between the devices therein. In the application of the solution, it can be set according to actual needs.

[0152] Next, the technical solution of the present application will be described in detail through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0153] Figure 2 is a schematic flowchart of the first embodiment of an intelligent question-answering method provided by the embodiments of the present application. Refer to Figure 2 and the method specifically includes the following steps:

[0154] S201: Obtain multilingual text.

[0155] In this embodiment, the computing device can obtain multilingual text. In one implementation, the computing device can obtain multilingual text sent by the terminal device. The multilingual text can be input by the user to the terminal device.

[0156] It should be noted that the multilingual text refers to text including multiple languages.

[0157] For example, the multilingual text can be "Our goal for this quarter is to complete this project. I hope everyone can ensure that all milestones are achieved on time before the deadline. If you have any questions, please call me in time."

[0158] In one implementation,

[0159] the computing device can obtain multilingual speech data.

[0160] The computing device can perform text recognition processing on the multilingual speech data to obtain multilingual text. In one implementation, the computing device can perform text recognition processing (speech-to-text conversion) on the multilingual speech data based on an Automatic Speech Recognition (ASR) model to obtain multilingual text.

[0161] S202: Performing language recognition and translation processing on the multilingual text to obtain a question text corresponding to the target language.

[0162] In this embodiment, the computing device can perform text language recognition and translation processing on the multilingual text to obtain the question text corresponding to the target language.

[0163] In one implementation,

[0164] The computing device can perform word segmentation processing on the multilingual text through the language recognition model to obtain multiple first keywords and determine the language corresponding to each of the multiple first keywords. For example, for the multilingual text "Excuse me, which lift can I take to the 5th floor?", the computing device can perform word segmentation processing on the multilingual text through the language recognition model and determine that the language corresponding to "Excuse me" is Chinese, the language corresponding to "sit" is Chinese, the language corresponding to "which" is Chinese, the language corresponding to "lift" is English, the language corresponding to "can" is Chinese, the language corresponding to "to" is Chinese, and the language corresponding to "5th floor" is Chinese. In one implementation, the computing device can also determine that the language corresponding to "?" is a symbol.

[0165] The computing device can translate the multilingual text according to the language corresponding to each first keyword and the target language using a language translation model to obtain the question text. For example, the target language can be Chinese.

[0166] Next, the process of a computing device acquiring a target language is described.

[0167] In one implementation,

[0168] The computing device may determine the language (Lan) corresponding to the largest number of first keywords among the multiple languages according to the languages corresponding to the first keywords. main ), which is the target language.

[0169] For example, for the multilingual text "Which elevator can I take to the 5th floor?", the computing device may determine that the number of first keywords corresponding to Chinese is 5, and the number of first keywords corresponding to English is 1. The computing device may determine that the language (Chinese) with the largest number of corresponding first keywords among the multiple languages is the target language.

[0170] In one implementation,

[0171] The computing device may determine, according to the languages corresponding to the first keywords, the languages in the multiple languages for which the number of corresponding first keywords is greater than the first threshold as the target languages. Exemplarily, the first threshold is 3. It should be noted that the first threshold is a preset threshold.

[0172] In one implementation,

[0173] The computing device may determine the preset language as the target language. It should be noted that the preset language is a pre-set language.

[0174] In one implementation,

[0175] The computing device may determine, according to the languages corresponding to the first keywords, the language in the multiple languages for which the number of corresponding first keywords is the largest as the candidate target language.

[0176] The computing device may, when determining that the number of candidate target languages is 1, determine the candidate target language as the target language. In one implementation, the computing device may, when determining that the number of candidate target languages is greater than 1, determine the preset language as the target language. In one implementation, the computing device may, when determining that the number of candidate target languages is greater than 1, determine any one of the multiple candidate target languages as the target language.

[0177] For example, for the multi-language text "Bonjour Mademoiselle, welcome to our store, sit down please", the computing device may determine that the number of first keywords corresponding to French is 2, the number of first keywords corresponding to Chinese is 3, and the number of first keywords corresponding to English is 3. The computing device may determine that both Chinese and English are the languages for which the number of corresponding first keywords is the largest. The computing device may determine Chinese and English as the candidate target languages. In one implementation, based on the number of candidate target languages being greater than 1, the computing device may determine the preset language (Chinese) as the target language. In one implementation, based on the number of candidate target languages being greater than 1, the computing device may determine any one of the two candidate target languages (such as English) as the target language.

[0178] In one implementation,

[0179] The computing device can determine the number of first keywords corresponding to each language according to the language to which each first keyword belongs. The computing device can determine the proportion of the first keywords corresponding to each language according to the number of the first keywords corresponding to each language. The computing device can determine the language with the largest proportion of the corresponding first keywords among multiple languages as the target language.

[0180] S203: Determine the answer text corresponding to the question text from the multilingual knowledge base according to the target language.

[0181] In this embodiment, the computing device can store a multilingual knowledge base. It should be noted that the multilingual knowledge base can include knowledge base versions corresponding to multiple languages. It should also be noted that the multilingual knowledge base is constructed according to preset knowledge points and large language models.

[0182] The computing device can determine the answer text corresponding to the question text from the multilingual knowledge base according to the target language.

[0183] Next, the process of the computing device determining the answer text corresponding to the question text from the multilingual knowledge base according to the target language will be described.

[0184] In one implementation,

[0185] The computing device can perform vectorization processing on the question text to obtain a first question vector (V Q ).

[0186] The computing device can query the multilingual knowledge base according to the first question vector and determine a target question vector from multiple initial question vectors. The multilingual knowledge base includes multiple initial question vectors and the initial answer text corresponding to each initial question vector. It should be noted that the multiple initial question vectors correspond to the target language. In one implementation, the multilingual knowledge base includes knowledge base versions corresponding to multiple languages. The computing device can determine the knowledge base version corresponding to the target language from the multilingual knowledge base according to the target language. The computing device can query the knowledge base version corresponding to the target language according to the first question vector and determine the target question vector from multiple initial question vectors. The knowledge base version corresponding to the target language includes multiple initial question vectors and the initial answer text corresponding to each initial question vector.

[0187] The computing device can determine the target answer text corresponding to the target question vector as the answer text.

[0188] Next, the process of the computing device querying the multilingual knowledge base according to the first question vector and determining the target question vector from multiple initial question vectors will be described.

[0189] In one implementation,

[0190] The computing device can determine the matching degree corresponding to each initial problem vector according to multiple initial problem vectors (V K [Lan main [qus]) and the first problem vector (V Q ).

[0191] In one implementation, the computing device can calculate the matching degree corresponding to the initial problem vector by calculating the cosine similarity. The calculation method of the cosine similarity is as follows:

[0192]

[0193] The computing device can determine the target problem vector from multiple initial problem vectors according to the matching degrees corresponding to the multiple initial problem vectors.

[0194] Next, the process of the computing device determining the target problem vector from multiple initial problem vectors according to the matching degrees corresponding to the multiple initial problem vectors will be described.

[0195] In one implementation,

[0196] The computing device can obtain the corresponding preset matching degree threshold according to the target language. It should be noted that the preset matching degree threshold is set in advance.

[0197] The computing device can determine the target problem vector from multiple initial problem vectors according to the matching degrees corresponding to the multiple initial problem vectors and the preset matching degree threshold.

[0198] Next, the process of the computing device determining the target problem vector from multiple initial problem vectors according to the matching degrees corresponding to the multiple initial problem vectors and the preset matching degree threshold will be described.

[0199] In one implementation,

[0200] The computing device can determine whether there is an initial problem vector in the multiple initial problem vectors whose corresponding matching degree is greater than or equal to the preset matching degree threshold.

[0201] If there is, the computing device can determine the initial problem vector whose corresponding matching degree is greater than or equal to the preset matching degree threshold as the candidate target problem vector. When the computing device determines that the number of candidate target problem vectors is 1, it can determine the candidate target problem vector as the target problem vector. When the computing device determines that the number of candidate target problem vectors is greater than 1, it can determine the candidate target problem vector with the largest corresponding matching degree among the multiple candidate target problem vectors as the target problem vector.

[0202] If not, the computing device can determine multiple candidate question vectors based on a preset number and the matching degrees corresponding to multiple initial question vectors. Exemplarily, when the preset number is 10, the computing device can determine the 10 initial question vectors with the highest matching degrees as the candidate question vectors. The computing device can obtain the candidate answer texts corresponding to each candidate question vector. The computing device can use a large language model to determine whether there is an answer text corresponding to the question text in the candidate answer texts corresponding to each candidate question vector. If so, the computing device can determine the candidate question vector corresponding to the candidate answer text as the target question vector. If not, the computing device can determine the answer text as a preset answer text. Exemplarily, the preset answer text can be "don't know".

[0203] Beneficial effects of this embodiment: In this embodiment, the computing device can obtain multilingual texts, perform text language identification and translation processing on the multilingual texts, and obtain a question text in the target language. The computing device can determine the answer text corresponding to the question text from a multilingual knowledge base according to the target language; wherein, the multilingual knowledge base is constructed according to preset knowledge points and a large language model. By the above method of performing text language identification and translation processing on multilingual texts, language alignment can be performed, enabling the computing device to fully understand the content of multilingual texts, and then retrieving the answer text from the multilingual knowledge base based on the question text in the target language, improving the accuracy of determining the answer text.

[0204] Figure 3a It is a schematic flowchart of the second embodiment of an intelligent question-answering method provided by an embodiment of the present application. Refer to Figure 3a and the method specifically includes the following steps:

[0205] S301: Obtain multilingual texts.

[0206] In this embodiment, the computing device can obtain multilingual texts.

[0207] The specific implementation process is the same as that of S201 and will not be elaborated here.

[0208] S302: Through a language identification model, perform word segmentation processing on the multilingual text to obtain multiple first keywords, and determine the language corresponding to each first keyword among the multiple first keywords.

[0209] In this embodiment, the computing device can perform word segmentation processing on the multilingual text through a language identification model to obtain multiple first keywords, and determine the language corresponding to each first keyword among the multiple first keywords.

[0210] Next, through specific examples, the process in which a computing device performs word segmentation on a multilingual text through a language recognition model to obtain multiple first keywords and determines the language corresponding to each first keyword among the multiple first keywords will be described.

[0211] Figure 3b This is a schematic diagram of a scenario for determining the language corresponding to each first keyword provided by an embodiment of the present application.

[0212] As Figure 3b shown, for the multilingual text "I hope everyone can ensure that all milestones are achieved on time before the deadline", the computing device can perform word segmentation on the multilingual text through a language recognition model to obtain multiple first keywords and determine the language corresponding to each keyword among the multiple first keywords: "hope <Chinese>", "everyone <Chinese>", "can <Chinese>", "before <Chinese>", "deadline <English>", "before <Chinese>", ",", "ensure <Chinese>", "all <Chinese>", "milestones <Chinese>", "are <English>", "can <Chinese>", "on time <Chinese>", "achieved <Chinese>".

[0213] For the multilingual text "I tried that new hot pot restaurant last night, and it was amazing", the computing device can perform word segmentation on the multilingual text through a language recognition model to obtain multiple first keywords and determine the language corresponding to each keyword among the multiple first keywords: "I <English>", "tried <English>", "that <English>", "new <English>", "hot pot restaurant <Chinese>", ",", "and <English>", "it <English>", "was <English>", "amazing <English>".

[0214] For the multilingual text "What is the size of this piece of clothing", the computing device can perform word segmentation on the multilingual text through a language recognition model to obtain multiple first keywords and determine the language corresponding to each keyword among the multiple first keywords: "this <Chinese>", "piece of clothing <Chinese>", "size <English>", "is <Chinese>", "what <Chinese>".

[0215] Next, the process in which the computing device trains an initial language recognition model to obtain a language recognition model will be described.

[0216] Figure 3c This is a schematic flowchart of training an initial language recognition model provided by an embodiment of the present application. As Figure 3c shown, the process of training a language recognition model includes the following steps:

[0217] S3c1: Obtain multiple text statements.

[0218] In this embodiment, the computing device may obtain the original corpus, where the original corpus includes multiple text statements. It should be noted that among the multiple text statements, there may be text statements in different languages. For a text statement, the text statement is a text statement in a single language.

[0219] Exemplarily, the computing device may collect texts such as datasets and encyclopedic information, take multiple text statements (such as 100,000 text statements), and construct the original corpus (multiple text statements).

[0220] S3c2: Perform word segmentation on the text statements to obtain multiple third keywords.

[0221] In this embodiment, the computing device may perform word segmentation on the text statements to obtain multiple third keywords.

[0222] S3c3: Perform translation processing on at least one of the third keywords in the text statement to obtain a sample text statement corresponding to the text statement.

[0223] In this embodiment, the computing device may perform translation processing on at least one of the third keywords in the text statement to obtain a sample text statement corresponding to the text statement.

[0224] It should be noted that in the sample text statement corresponding to the text statement, it includes at least one translated third keyword and at least one untranslated third keyword.

[0225] For example, for the text statement "What is the size of this dress", performing word segmentation on this text statement can obtain multiple third keywords "this", "of the dress", "size", "is", "what". The computing device may perform translation processing on the "size" in the text statement to obtain the sample text statement corresponding to the text statement "What is the size of this dress". In this sample text statement, it includes one translated third keyword "size" and four untranslated third keywords "this", "of the dress", "is", "what".

[0226] S3c4: Obtain the label information corresponding to the sample text statement.

[0227] In this embodiment, the computing device may obtain the label information corresponding to the sample text statement.

[0228] Among them, the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword.

[0229] Next, S3c1, S3c2, S3c3, and S3c4 will be described by way of example.

[0230] For example, a computing device may perform a partitioning process on multiple text statements to obtain a first set of text statements, a second set of text statements, and a third set of text statements. It should be noted that in a set of text statements, the languages corresponding to multiple text statements may be different. In one implementation, the number of text statements included in the first set of text statements, the number of text statements included in the second set of text statements, and the number of text statements included in the third set of text statements may be the same.

[0231] For any text statement in the first set of text statements, the computing device may perform a word segmentation process on the text statement to obtain multiple third keywords. The computing device may perform a translation process on any one of the multiple third keywords to obtain a sample text statement corresponding to the text statement. The computing device may obtain the label information corresponding to the sample text statement.

[0232] For any text statement in the second set of text statements, the computing device may perform a word segmentation process on the text statement to obtain multiple third keywords. The computing device may perform a translation process on any two of the multiple third keywords to obtain a sample text statement corresponding to the text statement. The computing device may obtain the label information corresponding to the sample text statement.

[0233] For any text statement in the third set of text statements, the computing device may determine the text statement as a sample text statement. The computing device may obtain the label information corresponding to the sample text statement.

[0234] S3c5: Train an initial language recognition model based on multiple sample text statements and the label information corresponding to each sample text statement to obtain a language recognition model.

[0235] In this embodiment, the computing device may train an initial language recognition model based on multiple sample text statements and the label information corresponding to each sample text statement to obtain a language recognition model.

[0236] In one implementation,

[0237] The computing device may construct a first training data set and a first test data set according to multiple sample text statements and the label information corresponding to each sample text statement. For example, the ratio of the number of sample text statements in the first training data set to the number of all sample text statements may be 3 / 4; the ratio of the number of sample text statements in the first test data set to the number of all sample text statements may be 1 / 4.

[0238] The computing device may train the initial language recognition model according to the first training data set to obtain a trained language recognition model. For example, the number of training iterations may be 1000 times.

[0239] The computing device may test the trained language recognition model according to the first test data set to obtain the accuracy rate of the trained language recognition model.

[0240] When the computing device determines that the accuracy rate of the trained language recognition model is greater than the first preset accuracy rate, it determines the trained language recognition model as the language recognition model. Exemplarily, the first preset accuracy rate may be 90%.

[0241] Figure 3d This is a schematic diagram of a model training - model testing scenario provided by an embodiment of the present application.

[0242] As Figure 3d shown, during the model training process,

[0243] The first training data set includes multiple sample text sentences and label information corresponding to each sample text sentence.

[0244] Exemplarily, Figure 3d shows a sample text sentence and the label information corresponding to this sample text sentence. Among them, the sample text sentence is "I hope everyone can ensure that all milestones are achieved on time before the deadline", and the label information includes "I hope <Chinese>", "everyone <Chinese>", "can <Chinese>", "before <Chinese>", "deadline <English>", "before <Chinese>", ",", "ensure <Chinese>", "all <Chinese>", "milestones <Chinese>", "are <Chinese>", "able to <Chinese>", "on time <Chinese>", "achieve <Chinese>". It should be noted that the sample text sentence is obtained by the computing device after performing word segmentation processing on the text sentence to obtain multiple third keywords and then performing translation processing on one of the third keywords (deadline) in the text sentence.

[0245] Figure 3dAnother sample text sentence and the corresponding tag information are shown. Among them, the sample text sentence is "I tried that new hot pot restaurant last night, and it was amazing", and the tag information includes "I<English>", "tried<English>", "that<English>", "new<English>", "hot pot restaurant<Chinese>", ",", "and<English>", "it<English>", "was<English>", "amazing<English>". It should be noted that the sample text sentence is obtained by the computing device after performing word segmentation on the text sentence to obtain multiple third keywords and then performing translation processing on one of the third keywords (hot pot restaurant) in the text sentence.

[0246] Figure 3d Another sample text sentence and the corresponding tag information are shown. Among them, the sample text sentence is "What is the size of this piece of clothing", and the tag information includes "this<Chinese>", "piece of clothing's<Chinese>", "size<English>", "is<Chinese>", "what<Chinese>". It should be noted that the sample text sentence is obtained by the computing device after performing word segmentation on the text sentence to obtain multiple third keywords and then performing translation processing on one of the third keywords (size) in the text sentence.

[0247] The computing device can train the initial language recognition model according to the first training data set to obtain the trained language recognition model.

[0248] During the model testing process,

[0249] The first test data set includes multiple sample text sentences and the corresponding tag information for each sample text sentence.

[0250] Exemplarily, Figure 3d A sample text sentence and the corresponding tag information are shown. Among them, the sample text sentence is "I hope everyone can ensure that all milestones are achieved on time before the deadline", and the tag information includes "hope<Chinese>", "everyone<Chinese>", "can<Chinese>", "before<Chinese>", "deadline<Chinese>", ",", "ensure<Chinese>", "all<Chinese>", "milestones<English>", "are<Chinese>", "achieved<Chinese>". It should be noted that the sample text sentence is obtained by the computing device after performing word segmentation on the text sentence to obtain multiple third keywords and then performing translation processing on one of the third keywords (milestone) in the text sentence.

[0251] Figure 3e Another sample text sentence and the tag information corresponding to the sample text sentence are shown. The sample text sentence is "I tried that new hot pot last night, and it was awesome," and the tag information includes "I <English>," "tried <English>," "that <English>," "new <English>," "hot pot <English>," ", <symbol>," "and <English>," "it <English>," "was <English>," and "amazing <Chinese>." It should be noted that the sample text sentence is obtained by translating one of the third keywords (amazing) in the text sentence after the computing device performs word segmentation processing on the text sentence to obtain multiple third keywords.

[0252] The computing device may process the sample text sentences in the first test dataset using the trained language identification model to obtain output label information. In one implementation, the computing device may perform word segmentation processing on the sample text sentences using the trained language identification model to obtain multiple first output keywords, determine the language corresponding to each of the multiple first output keywords, and determine that the output label information includes the language corresponding to each of the first output keywords.

[0253] The computing device may determine an accuracy rate of the trained language recognition model based on the output label information and the label information in the first test data set. If the computing device determines that the accuracy rate is greater than a first preset accuracy rate, the computing device may determine the trained language recognition model as the language recognition model.

[0254] S303: According to the languages corresponding to the first keywords and the target language, the multilingual text is translated using a language translation model to obtain a question text.

[0255] In this embodiment, the computing device may translate the multilingual text according to the language corresponding to each first keyword and the target language through a language translation model to obtain the question text.

[0256] The following describes, through a specific example, a process in which a computing device translates a multilingual text according to the language corresponding to each first keyword and the target language using a language translation model to obtain a question text.

[0257] Figure 3e A schematic diagram of a scenario for obtaining question text provided in an embodiment of the present application.

[0258] like Figure 3eAs shown, the computing device can translate the multilingual text "I hope everyone can ensure that all milestones can be achieved on time before the deadline" through the first translation model according to the language corresponding to each first keyword ("hope <Chinese>", "everyone <Chinese>", "can <Chinese>", "in <Chinese>", "deadline <English>", "before <Chinese>", ", <symbol>", "ensure <Chinese>", "all <Chinese>", "milestone <Chinese>", "all <Chinese>", "can <Chinese>", "on time <Chinese>", "achieve <Chinese>") and the target language (Chinese), and obtain the question text "I hope everyone can ensure that all milestones can be achieved on time before the deadline".

[0259] The computing device can translate the multilingual text "I tried that new hot pot restaurant last night, and it was amazing" through the first translation model according to the language corresponding to each first keyword ("I<English>", "tried<English>", "that<English>", "new<English>", "hot pot restaurant<Chinese>", ", <symbol>", "and<English>", "it<English>", "was<English>", "amazing<English>") and the target language (English), and obtain the question text "Itried that new hot pot last night, and it was amazing".

[0260] The computing device can translate the multilingual text "What is the size of this dress" through the first translation model according to the language corresponding to each first keyword ("this <Chinese>", "the <Chinese> of the dress", "size <English>", "is <Chinese>", "how big <Chinese>") and the target language (Chinese), and obtain the question text "What is the size of this dress".

[0261] Next, the process of training the initial language translation model on a computing device to obtain the language translation model is described.

[0262] Figure 3f A flow chart of a training language translation model provided in an embodiment of the present application. Figure 3f As shown in the figure, the process of training a language translation model includes the following steps:

[0263] S3f1: Get multiple text sentences and the languages corresponding to the text sentences.

[0264] In this embodiment, the computing device may obtain a plurality of text sentences and the languages corresponding to the text sentences.

[0265] It should be noted that the process by which the computing device obtains multiple text statements is the same as that of S3c1 and will not be elaborated here.

[0266] S3f2: Perform word segmentation on the text statement to obtain multiple third keywords.

[0267] In this embodiment, the computing device can perform word segmentation on the text statement to obtain multiple third keywords.

[0268] The specific implementation process is the same as that of S3c2 and will not be elaborated here.

[0269] S3f3: Perform translation processing on at least one third keyword of the text statement to obtain a sample text statement corresponding to the text statement.

[0270] In this embodiment, the computing device can perform translation processing on at least one third keyword of the text statement to obtain a sample text statement corresponding to the text statement.

[0271] The specific implementation process is the same as that of S3c3 and will not be elaborated here.

[0272] S3f4: Obtain the label information corresponding to the sample text statement.

[0273] In this embodiment, the computing device can obtain the label information corresponding to the sample text statement.

[0274] Among them, the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword.

[0275] The specific implementation process is the same as that of S3c4 and will not be elaborated here.

[0276] S3f5: According to multiple text statements, the languages corresponding to each text statement, the sample text statements, and the label information corresponding to each sample text statement, perform training processing on the initial language translation model to obtain a language translation model.

[0277] In this embodiment, the computing device can perform training processing on the initial language translation model according to multiple text statements, the languages corresponding to each text statement, the sample text statements, and the label information corresponding to each sample text statement to obtain a language translation model.

[0278] The computing device can construct a second training dataset and a second test dataset based on multiple text statements, the language corresponding to each text statement and the sample text statement, and the label information corresponding to each sample text statement. For example, the ratio of the number of text statements in the second training dataset to the number of all text statements can be 3 / 4; the ratio of the number of text statements in the second test dataset to the number of all text statements can be 1 / 4.

[0279] The computing device can adjust the initial language translation model based on the second training dataset to obtain a trained language translation model. For example, the number of training iterations can be 1000 times. For example, the initial language translation model can be a Transformer model; for another example, the initial language translation model can be a Bidirectional Encoder Representations from Transformers (BERT) model based on Transformer; for another example, the initial language translation model can be an mBART model.

[0280] The computing device can test the trained language translation model based on the second test dataset to obtain the accuracy of the trained language translation model. When the computing device determines that the accuracy of the trained language translation model is greater than the second preset accuracy, the trained language translation model is determined as the language translation model. Exemplarily, the second preset accuracy can be 90%.

[0281] Figure 3g This is another scenario schematic diagram of model training - model testing provided by the embodiments of this application.

[0282] As Figure 3g shown, during the model training process,

[0283] The second training dataset can include multiple text statements, the language corresponding to each text statement and the sample text statement, and the label information corresponding to each sample text statement.

[0284] Exemplarily, Figure 3gShows a text statement, the language corresponding to the text statement and a sample text statement, and the tag information corresponding to the sample text statement. Among them, the text statement is "I hope everyone can ensure that all milestones are achieved on time before the deadline", the language corresponding to the text statement is Chinese, the sample text statement corresponding to the text statement is "I hope everyone can ensure that all milestones are achieved on time before the deadline", and the tag information corresponding to the sample text statement includes "I hope <Chinese>", "everyone <Chinese>", "can <Chinese>", "before <Chinese>", "deadline <English>", "before <Chinese>", ",", "ensure <Chinese>", "all <Chinese>", "milestones <Chinese>", "can <Chinese>", "on time <Chinese>", "achieve <Chinese>". It should be noted that the sample text statement is obtained by the computing device after performing word segmentation on the text statement to obtain multiple third keywords and then translating one of the third keywords (deadline) in the text statement.

[0285] Exemplarily, Figure 3g Shows another text statement, the language corresponding to the text statement and a sample text statement, and the tag information corresponding to the sample text statement. Among them, the text statement is "I tried that new hot potlast night, and it was amazing", the language corresponding to the text statement is Chinese, the sample text statement corresponding to the text statement is "I tried that new hot pot restaurant last night, and it was amazing", and the tag information corresponding to the sample text statement includes "I <English>", "tried <English>", "that <English>", "new <English>", "hot pot restaurant <Chinese>", ",", "and <English>", "it <English>", "was <English>", "amazing <English>". It should be noted that the sample text statement is obtained by the computing device after performing word segmentation on the text statement to obtain multiple third keywords and then translating one of the third keywords (hot pot restaurant) in the text statement.

[0286] Exemplarily, Figure 3gShows another text statement, the language corresponding to the text statement and the sample text statement, and the label information corresponding to the sample text statement. Among them, the text statement is "What is the size of this piece of clothing", the language corresponding to the text statement is Chinese, the sample text statement corresponding to the text statement is "What is the size of this piece of clothing", and the label information corresponding to the sample text statement includes "<Chinese> for this", "<Chinese> for the clothing", "<English> for size", "<Chinese> for is", and "<Chinese> for how big". It should be noted that the sample text statement is obtained by the computing device after performing word segmentation on the text statement to obtain multiple third keywords, and then performing translation processing on one of the third keywords (size) in the text statement.

[0287] The computing device can train the initial language translation model according to the second training dataset to obtain the trained language translation model.

[0288] During the model testing process,

[0289] The second training dataset can include multiple text statements, the language corresponding to each text statement and the sample text statement, and the label information corresponding to each sample text statement.

[0290] Exemplarily, Figure 3g Shows a sample text statement "I hope everyone can ensure that all milestones are achieved on time before the deadline". It should be noted that the text statement corresponding to this sample text statement ( Figure 3g not shown) is "I hope everyone can ensure that all milestones are achieved on time before the deadline", the language corresponding to this text statement ( Figure 3g not shown) is Chinese, and the label information corresponding to this sample text statement ( Figure 3g not shown) includes "<Chinese> for hope", "<Chinese> for everyone", "<Chinese> for can", "<Chinese> for at", "<Chinese> for deadline", "<Chinese> for before", "<symbol> for,", "<Chinese> for ensure", "<Chinese> for all", "<English> for milestone", "<Chinese> for all", "<Chinese> for can", "<Chinese> for on time", and "<Chinese> for achieve".

[0291] Exemplarily, Figure 3g Shows another sample text statement "I tried that new hot pot lastnight, and it was so cool". It should be noted that the text statement corresponding to this sample text statement ( Figure 3gnot shown) is "I tried that new hot pot last night, and it was amazing", and the language corresponding to this text statement Figure 3g not shown) is English, and the label information corresponding to this sample text statement Figure 3g not shown) includes "I<English>", "tried<English>", "that<English>", "new<English>", "hot pot<English>", ",", "and<English>", "it<English>", "was<English>", "so cool<Chinese>".

[0292] The computing device can process the sample text statements in the second test dataset through the trained language translation model according to the language corresponding to the text statements in the second test dataset and the label information corresponding to the sample text statements, to obtain output text statements. The computing device can determine the accuracy rate of the trained language translation model according to the output text statements and the text statements corresponding to the sample text statements in the second test dataset. The computing device can, when determining that the accuracy rate is greater than the second preset accuracy rate, determine the trained language translation model as the language translation model.

[0293] S304: Determine the answer text corresponding to the question text from the multilingual knowledge base according to the target language.

[0294] In this embodiment, the computing device can determine the answer text corresponding to the question text from the multilingual knowledge base according to the target language.

[0295] Beneficial effects of this embodiment: In this embodiment, the computing device can obtain multilingual texts. The computing device can perform word segmentation processing on the multilingual texts through the language recognition model to obtain multiple first keywords, and determine the language corresponding to each first keyword among the multiple first keywords. The computing device can perform translation processing on the multilingual texts through the language translation model according to the language corresponding to each first keyword and the target language to obtain the question text. The computing device can determine the answer text corresponding to the question text from the multilingual knowledge base according to the target language; wherein, the multilingual knowledge base is constructed according to preset knowledge points and large language models. By the above method of determining the language corresponding to each first keyword among the multiple first keywords included in the multilingual text, and performing translation processing on the multilingual text according to the language corresponding to each first keyword and the target language, the languages can be aligned, so that the computing device can fully understand the content of the multilingual text, and then retrieve the multilingual knowledge base based on the question text corresponding to the target language to obtain the answer text, improving the accuracy rate of determining the answer text.

[0296] Figure 4Schematic flowchart of Embodiment 3 of an intelligent question-answering method provided by an embodiment of this application. Refer to Figure 4 , the method specifically includes the following steps:

[0297] S401: Obtain multilingual text.

[0298] In this embodiment, the computing device can obtain multilingual text.

[0299] In one implementation, the computing device can obtain multilingual speech data. The computing device can perform text recognition processing on the multilingual speech data to obtain multilingual text.

[0300] The specific implementation process is the same as that of S201 and will not be elaborated here.

[0301] S402: Perform text language identification and translation processing on the multilingual text to obtain the question text corresponding to the target language.

[0302] In this embodiment, the computing device can perform text language identification and translation processing on the multilingual text to obtain the question text corresponding to the target language.

[0303] The specific implementation process is the same as that of S202 and will not be elaborated here.

[0304] S403: According to the target language, determine the answer text corresponding to the question text from the multilingual knowledge base.

[0305] In this embodiment, the computing device can determine the answer text corresponding to the question text from the multilingual knowledge base according to the target language.

[0306] Among them, the multilingual knowledge base is constructed according to preset knowledge points and large language models.

[0307] The specific implementation process is the same as that of S203 and will not be elaborated here.

[0308] S404: Perform word segmentation processing on the answer text to obtain multiple second keywords.

[0309] In this embodiment, the computing device can perform word segmentation processing on the answer text to obtain multiple second keywords.

[0310] For example, for the answer text "Taking the fifth elevator can reach the 5th floor", the computing device can perform word segmentation processing on the answer text to obtain multiple second keywords "Taking", "fifth", "elevator", "can", "reach", "5th floor".

[0311] S405: Perform translation processing on at least one second keyword in the answer text to obtain a multilingual answer text.

[0312] In this embodiment, the computing device may perform translation processing on at least one second keyword in the answer text to obtain a multilingual answer text.

[0313] In one implementation,

[0314] The computing device may determine a second keyword that has an association relationship with the first keyword as the target second keyword. In one implementation, a second keyword that has an association relationship with the first keyword refers to a second keyword that has the same content as the first keyword but a different language. For example, based on a second keyword "elevator" in the second answer text, which has the same content as the first keyword "lift" but a different language, the computing device may determine this second keyword as the target keyword.

[0315] The computing device may determine the language corresponding to the first keyword as the language corresponding to the target second keyword.

[0316] The computing device may perform translation processing on the target second keyword in the answer text according to the language corresponding to the target second keyword to obtain a multilingual answer text. For example, the multilingual answer text may be "You can take the fifth lift to the 5th floor".

[0317] In addition, in one implementation, after obtaining the multilingual answer text, the computing device may generate a voice Q&A result according to the multilingual answer text. In one implementation, the computing device may send the voice Q&A result to the terminal device for the terminal device to perform playback processing on the voice Q&A result.

[0318] Advantages of this embodiment: The computing device can obtain a multilingual text, perform text language recognition and translation processing on the multilingual text to obtain a question text corresponding to the target language. The computing device may determine an answer text corresponding to the question text from a multilingual knowledge base according to the target language; wherein, the multilingual knowledge base is constructed based on preset knowledge points and a large language model. The computing device may perform word segmentation processing on the answer text to obtain multiple second keywords, and perform translation processing on at least one second keyword in the answer text to obtain a multilingual answer text. On the one hand, through the above-mentioned method of performing text language recognition and translation processing on the multilingual text, language alignment can be performed, enabling the computing device to fully understand the content of the multilingual text, and then retrieving the answer text from the multilingual knowledge base based on the question text corresponding to the target language, improving the accuracy of determining the answer text. On the other hand, by converting the answer text into a multilingual answer text, the personalized Q&A needs of users can be met, improving the user's Q&A experience.

[0319] Figure 5Schematic flowchart of Embodiment 4 of an intelligent question-answering method provided by an embodiment of this application. Refer to Figure 5 , the method specifically includes the following steps:

[0320] S501: Obtain initial library versions corresponding to multiple languages.

[0321] In this embodiment, the computing device can obtain initial library versions corresponding to multiple languages.

[0322] Among them, the initial library corresponding to one language includes multiple initial question texts and initial answer texts corresponding to each initial question text.

[0323] Next, the process of the computing device obtaining initial library versions corresponding to multiple languages will be described.

[0324] In one implementation,

[0325] The computing device can obtain preset knowledge points; among them, the preset knowledge points correspond to the first language. In one implementation, the computing device can obtain an original knowledge base. The original knowledge base includes preset knowledge points. For example, the preset knowledge points can include "buying high-speed rail tickets on a ticketing software", "adjusting the wind speed of an air conditioner using an air conditioner remote control", "buying a refrigerator on a shopping software", and so on.

[0326] The computing device can obtain the initial library version corresponding to the first language according to the preset knowledge points and a large language model. It should be noted that the initial question texts and initial answer texts in the initial library version corresponding to the first language correspond to the first language. Exemplarily, the first language can be Chinese. For example, the initial question text can be "How to buy train tickets", and the initial answer text can be "Buy train tickets on a ticketing software".

[0327] In one implementation, the computing device can obtain prompt information. Among them, the prompt information includes preset knowledge points. The computing device can control the large language model according to the prompt information to generate the initial library version corresponding to the first language. Exemplarily, the prompt information can be "You are a question-answering system. I will provide you with preset knowledge points in the original knowledge base. You need to generate a series of corresponding initial question texts based on the preset knowledge points. The same preset knowledge point can be asked in different ways, and you can output the results of multiple initial question text - initial answer text pairs. The same initial answer text should be output with at least 3 different initial question texts. The preset knowledge points in the original knowledge base are as follows: "XXX". Please output the initial question text - initial answer text according to the requirements. Note that the language of the initial question text - initial answer text needs to be the same as the language of the preset knowledge points in the original knowledge base".

[0328] The computing device can perform translation processing on the initial library version corresponding to the first language to obtain the initial library version corresponding to the second language. Among them, the initial question text and the initial answer text in the initial library version corresponding to the second language correspond to the second language. In one implementation, the computing device can use a translation model to perform translation processing on the initial library version corresponding to the first language to obtain the initial library corresponding to the second language. For example, the translation model can be a Transformer model; for another example, the translation model can be a Bidirectional Encoder Representations from Transformers (BERT) model based on Transformer; for another example, the translation model can be an mBART model.

[0329] Exemplarily, the second language can be Chinese, French, Spanish, etc.

[0330] S502: Vectorize the initial question text in multiple initial library versions to obtain knowledge base versions corresponding to multiple languages.

[0331] In this embodiment, the computing device can vectorize the initial question text in multiple initial library versions to obtain knowledge base versions corresponding to multiple languages. It should be noted that vectorization processing means converting the initial question text into a vector.

[0332] Among them, the knowledge base version (V K [Lan]) corresponding to a language (Lan) includes multiple initial question vectors (qus) and the initial answer text (V K [Lan][qus]) corresponding to each initial question vector.

[0333] In one implementation, the computing device can use Word2Vec to vectorize the initial question text in multiple initial library versions to obtain knowledge base versions corresponding to multiple languages.

[0334] In one implementation, the computing device can use BERT to vectorize the initial question text in multiple initial library versions to obtain knowledge base versions corresponding to multiple languages.

[0335] S503: Construct a multilingual knowledge base according to the knowledge base versions corresponding to multiple languages.

[0336] In this embodiment, the computing device can construct a multilingual knowledge base according to the knowledge base versions corresponding to multiple languages.

[0337] Among them, the multilingual knowledge base includes knowledge base versions corresponding to multiple languages.

[0338] Beneficial effects of this embodiment: In this embodiment, the computing device can obtain the initial library versions corresponding to multiple languages. Among them, the initial library includes multiple initial question texts and the corresponding initial answer texts for each initial question text. The computing device can perform vectorization processing on the initial question texts in multiple initial library versions to obtain the knowledge base versions corresponding to multiple languages. The computing device can construct a multilingual knowledge base based on the knowledge base versions corresponding to multiple languages. By pre-constructing the multilingual knowledge base, in scenarios where intelligent question answering for multilingual texts is required, the answer text corresponding to the question text can be quickly determined from the multilingual knowledge base according to the target language. That is to say, by pre-constructing the multilingual knowledge base, different language question-and-answer scenarios can be better handled, and the query accuracy of the answer text is improved.

[0339] The following is an embodiment of the apparatus of the present application, which can be used to execute the method embodiment of the present application. For details not disclosed in the apparatus embodiment of the present application, please refer to the method embodiment of the present application.

[0340] Figure 6 It is a schematic structural diagram of an intelligent question answering device provided by an embodiment of the present application. The intelligent question answering device 60 is applied to a computing device.

[0341] As Figure 6 shown, the intelligent question answering device 60 includes an acquisition module 61 and a processing module 62. Among them,

[0342] The acquisition module 61 is used to acquire multilingual texts;

[0343] The processing module 62 is used to perform text language recognition and translation processing on the multilingual text to obtain the question text corresponding to the target language;

[0344] The processing module 62 is further used to determine the answer text corresponding to the question text from the multilingual knowledge base according to the target language; among them, the multilingual knowledge base is constructed based on preset knowledge points and large language models.

[0345] The intelligent question answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiment, and its beneficial effects are similar, so details will not be described here again.

[0346] In one implementation, the processing module 62 is specifically used for:

[0347] Perform word segmentation processing on the multilingual text through a language recognition model to obtain multiple first keywords, and determine the language corresponding to each first keyword in the multiple first keywords;

[0348] According to the languages corresponding to each first keyword and the target language, the multi-language text is translated through a language translation model to obtain a problem text.

[0349] The intelligent question-answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details will not be described herein again.

[0350] In one implementation manner, the processing module 62 is further configured to:

[0351] According to the languages corresponding to each first keyword, determine the language with the largest number of corresponding first keywords among the multiple languages as the target language; or,

[0352] According to the languages corresponding to each first keyword, determine the languages among the multiple languages in which the number of corresponding first keywords is greater than a first threshold as the target language; or,

[0353] Determine a preset language as the target language.

[0354] The intelligent question-answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details will not be described herein again.

[0355] In one implementation manner, the processing module 62 is specifically configured to:

[0356] Perform vectorization processing on the problem text to obtain a first problem vector;

[0357] According to the first problem vector, query a multi-language knowledge base, and determine a target problem vector from multiple initial problem vectors; the multi-language knowledge base includes multiple initial problem vectors and initial answer texts corresponding to each initial problem vector; wherein, the multiple initial problem vectors correspond to the target language;

[0358] Determine the target answer text corresponding to the target problem vector as the answer text.

[0359] The intelligent question-answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details will not be described herein again.

[0360] In one implementation manner, the processing module 62 is specifically configured to:

[0361] Determine the matching degree corresponding to each initial problem vector according to the multiple initial problem vectors and the first problem vector;

[0362] Determine a target problem vector from the multiple initial problem vectors according to the matching degrees corresponding to the multiple initial problem vectors.

[0363] The intelligent question answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0364] In one implementation, the processing module 62 is specifically configured to:

[0365] Obtain a corresponding preset matching degree threshold according to the target language;

[0366] Determine a target question vector from multiple initial question vectors according to the matching degrees of the multiple initial question vectors and the preset matching degree threshold.

[0367] The intelligent question answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0368] In one implementation, the processing module 62 is further configured to:

[0369] Perform word segmentation on the answer text to obtain multiple second keywords;

[0370] Perform translation processing on at least one of the second keywords in the answer text to obtain a multi-language answer text.

[0371] The intelligent question answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0372] In one implementation, the processing module 62 is further configured to:

[0373] Obtain multiple text sentences;

[0374] Perform word segmentation on the text sentences to obtain multiple third keywords;

[0375] Perform translation processing on at least one of the third keywords in the text sentences to obtain a sample text sentence corresponding to the text sentence;

[0376] Obtain label information corresponding to the sample text sentence; the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword;

[0377] Train the initial language recognition model according to the multiple sample text sentences and the label information corresponding to each sample text sentence to obtain a language recognition model.

[0378] The intelligent question answering device provided by the embodiment of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0379] In one implementation, the processing module 62 is further configured to:

[0380] Obtain multiple text statements and the languages corresponding to the text statements;

[0381] Perform word segmentation on the text statements to obtain multiple third keywords;

[0382] Perform translation processing on at least one of the third keywords in the text statements to obtain a sample text statement corresponding to the text statement;

[0383] Obtain the label information corresponding to the sample text statement; the label information includes the languages corresponding to at least one translated third keyword and the languages corresponding to at least one untranslated third keyword;

[0384] Train the initial language translation model according to the multiple text statements, the languages corresponding to each text statement, the sample text statements, and the label information corresponding to each sample text statement to obtain a language translation model.

[0385] The intelligent question-answering device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0386] In one implementation, the processing module 62 is further configured to:

[0387] Obtain the initial library versions corresponding to multiple languages; the initial library versions include multiple initial question texts and the initial answer texts corresponding to each initial question text;

[0388] Perform vectorization processing on the initial question texts in the initial libraries corresponding to multiple languages to obtain the knowledge base versions corresponding to multiple languages; the knowledge base corresponding to a language includes multiple initial question vectors and the initial answer texts corresponding to each initial question vector;

[0389] Construct a multilingual knowledge base according to the knowledge base versions corresponding to multiple languages.

[0390] The intelligent question-answering device provided by the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0391] In one implementation, the processing module 62 is specifically configured to:

[0392] Obtain a preset knowledge point; wherein, the preset knowledge point corresponds to the first language;

[0393] Obtain the initial library version corresponding to the first language according to the preset knowledge point and the large language model; the initial question texts and the initial answer texts in the initial library version corresponding to the first language correspond to the first language;

[0394] Perform translation processing on the initial library version corresponding to the first language to obtain the initial library version corresponding to the second language; the initial question text and the initial answer text in the initial library version corresponding to the second language correspond to the second language.

[0395] The intelligent question-answering device provided in the embodiments of the present application can execute the technical solutions in the above method embodiments, and the beneficial effects are similar, so details are not described herein again.

[0396] Figure 7 It is a schematic structural diagram of a computing device provided in the embodiments of the present application. As Figure 7 shown, the computing device 70 includes: a processor 71 and a memory 72; wherein, the processor 71 is coupled to the memory 72, and the memory 72 is used to store computer instructions; the processor 71 is used to execute the computer instructions to enable the computing device 70 to execute the technical solutions in the foregoing method embodiments.

[0397] Optionally, the memory 72 can be either independent or integrated with the processor 71. Optionally, when the memory 72 is a device independent of the processor 71, the computing device 70 can further include: a bus 73 for connecting the above devices.

[0398] The processor is used to execute the technical solutions in the foregoing method embodiments, and the implementation principles and technical effects are similar, so details are not described herein again.

[0399] The embodiments of the present application provide a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, they are used to implement the technical solutions provided in the foregoing method embodiments.

[0400] The embodiments of the present application provide a computer program product, including a computer program, and when the computer program is executed by a processor, it is used to implement the technical solutions provided in the foregoing method embodiments.

[0401] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media such as volatile memory and non-volatile memory that can store program codes.

[0402] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent question-answering method, characterized in that, The method includes: Obtain multilingual texts; Perform text language identification and translation processing on the multilingual texts to obtain a problem text corresponding to the target language; According to the target language, determine an answer text corresponding to the problem text from a multilingual knowledge base; wherein, the multilingual knowledge base is constructed based on preset knowledge points and a large language model.

2. The method according to claim 1, characterized in that, The performing text language identification and translation processing on the multilingual texts to obtain a problem text corresponding to the target language includes: Perform word segmentation processing on the multilingual texts through a language identification model to obtain a plurality of first keywords, and determine the language corresponding to each first keyword among the plurality of first keywords; According to the languages corresponding to the respective first keywords and the target language, perform translation processing on the multilingual texts through a language translation model to obtain the problem text.

3. The method according to claim 2, characterized in that, The method further includes: Determine, as the target language, the language with the largest number of corresponding first keywords among the plurality of languages according to the languages corresponding to the respective first keywords; or, Determine, as the target language, the language in which the number of corresponding first keywords among the plurality of languages is greater than a first threshold according to the languages corresponding to the respective first keywords; or, Determine a preset language as the target language.

4. The method according to any one of claims 1 to 3, characterized in that, The determining an answer text corresponding to the problem text from a multilingual knowledge base according to the target language includes: Perform vectorization processing on the problem text to obtain a first problem vector; According to the first problem vector, query the multilingual knowledge base and determine a target problem vector from a plurality of initial problem vectors; the multilingual knowledge base includes the plurality of initial problem vectors and initial answer texts corresponding to each of the initial problem vectors; wherein, the plurality of initial problem vectors correspond to the target language; Determine the target answer text corresponding to the target problem vector as the answer text.

5. The method according to claim 4, wherein The querying the multilingual knowledge base according to the first problem vector and determining a target problem vector from a plurality of initial problem vectors includes: Determine the matching degree corresponding to each initial problem vector according to the plurality of initial problem vectors and the first problem vector; Determine the target problem vector from the plurality of initial problem vectors according to the matching degrees corresponding to the plurality of initial problem vectors.

6. The method according to claim 5, characterized in that, The determining the target problem vector from the plurality of initial problem vectors according to the matching degrees corresponding to the plurality of initial problem vectors includes: Obtain a corresponding preset matching degree threshold according to the target language; Determine the target problem vector from the plurality of initial problem vectors according to the matching degrees corresponding to the plurality of initial problem vectors and the preset matching degree threshold.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: Perform word segmentation processing on the answer text to obtain a plurality of second keywords; Perform translation processing on at least one of the second keywords in the answer text to obtain a multilingual answer text.

8. The method according to claim 2, characterized in that The method further includes: Obtain a plurality of text sentences; Perform word segmentation processing on the text sentences to obtain a plurality of third keywords; Perform translation processing on at least one third keyword in the text statement to obtain a sample text statement corresponding to the text statement; Obtain the tag information corresponding to the sample text statement; the tag information includes the language corresponding to at least one translated third keyword and the language corresponding to at least one untranslated third keyword; Perform training processing on the initial language recognition model according to multiple sample text statements and the tag information corresponding to each sample text statement to obtain the language recognition model.

9. The method according to claim 2, wherein The method further includes: Obtain multiple text statements and the languages corresponding to the text statements; Perform word segmentation processing on the text statement to obtain multiple third keywords; Perform translation processing on at least one third keyword in the text statement to obtain a sample text statement corresponding to the text statement; Obtain the tag information corresponding to the sample text statement; the tag information includes the language corresponding to at least one translated third keyword and the language corresponding to at least one untranslated third keyword; Perform training processing on the initial language translation model according to multiple text statements, the languages corresponding to each text statement and the sample text statements, and the tag information corresponding to each sample text statement to obtain the language translation model.

10. A computing device, characterized in that, Includes: A processor and a memory communicatively connected to the processor; The memory is used to store computer execution instructions; The processor is used to execute the computer execution instructions stored in the memory to implement the method according to any one of claims 1-9.

Images