Interaction method and device, and computer readable storage medium

Abstract

An interaction method, an interaction device, and a computer readable storage medium, relating to the technical field of computers. The interaction method comprises: in response to reception of an input question of a user, searching on the basis of the input question to obtain a search result (S1); determining a search intent complexity level of the input question, different search intent complexity levels corresponding to different first machine learning models (S2); and on the basis of a first machine learning model corresponding to the search intent complexity level of the input question, processing the search result and generating an answer to the input question, wherein processing modes corresponding to different search intent complex levels are different (S3).

Description

Interaction method, device and computer readable storage medium TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, and particularly relates to an interaction method, device and computer readable storage medium. BACKGROUND

[0002] Human-computer interaction technology is an important research content in the field of artificial intelligence, and is a new interaction way between human and computer. With the gradual maturity of machine learning technology, the interaction technology has made significant progress and has new development prospects.

[0003] Dialogue technology is a main research direction of human-computer interaction technology. The dialogue interaction technology makes the computer able to understand and respond to the language input of the user by simulating the dialogue way of human. With the increase of the user amount, more and more demands gradually emerge. In the related technology, after the user inputs a question, a machine learning model is used to answer the question of the user. Usually, all the questions are answered by using the same model.

[0004] SUMMARY

[0005] In the related technology, when the dialogue interaction with the user is carried out, all the questions are answered by using the same model, and the accuracy of the generated answer and the interaction efficiency are not high.

[0006] In view of this, the embodiments of the present disclosure provide an interaction method, device and computer readable storage medium, which improve the accuracy of the answer and the interaction efficiency.

[0007] According to a first aspect of some embodiments of the present disclosure, an interaction method is provided, comprising: in response to receiving an input question of a user, searching according to the input question to obtain a search result; determining a search intention complexity level of the input question, wherein different search intention complexity levels correspond to different first machine learning models; processing the search result based on the first machine learning model corresponding to the search intention complexity level of the input question to generate an answer to the input question, wherein different search intention complexity levels correspond to different processing manners.

[0008] According to a second aspect of some embodiments of the present disclosure, there is provided an interactive device, comprising: a searching module configured to, in response to receiving an input question of a user, search according to the input question to obtain a search result; a determining module configured to determine a search intention complexity level of the input question, wherein different search intention complexity levels correspond to different first machine learning models; and a generating module configured to process the search result based on the first machine learning model corresponding to the search intention complexity level of the input question to generate an answer to the input question, wherein different search intention complexity levels correspond to different processing manners.

[0009] According to a third aspect of some embodiments of the present disclosure, there is provided an interactive device, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute an interactive method according to any of the embodiments of the present disclosure based on instructions stored in the memory.

[0010] According to a fourth aspect of some embodiments of the present disclosure, there is provided a computer readable storage medium having computer program instructions stored thereon, the instructions being executable by a processor to implement an interactive method according to any of the embodiments of the present disclosure.

[0011] According to a fifth aspect of some embodiments of the present disclosure, there is provided a computer program product comprising computer program instructions, the computer program instructions being executable by a processor to implement an interactive method according to any of the embodiments of the present disclosure.

[0012] This summary is provided to introduce some concepts of the present disclosure in a simplified form that are further described below in the detailed description section. This summary is not intended to identify key features or essential features of the claimed technology, nor is it intended to limit the scope of the claimed technology.

[0013] Other features of the present disclosure, which are not described in the detailed description section, will become apparent from the detailed description section and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0014] Preferred embodiments of the present disclosure are described herein with reference to the accompanying drawings. The drawings are provided to further enable those skilled in the art to practice the present disclosure, and are not intended to limit the scope of the present disclosure. In the drawings, like reference numerals refer to like elements throughout.

[0015] FIG. 1 shows a flowchart of an interactive method according to some embodiments of the present disclosure;

[0016] FIG. 2 shows a flowchart of an interaction method according to some embodiments of the present disclosure;

[0017] FIG. 3 shows a flowchart of searching according to an input question according to some embodiments of the present disclosure;

[0018] FIG. 4 shows a flowchart of an interaction method according to some embodiments of the present disclosure;

[0019] FIG. 5 shows a flowchart of determining a search intention complexity level of an input question according to some embodiments of the present disclosure;

[0020] FIGS. 6A-6B show flowcharts of displaying an interface according to some embodiments of the present disclosure;

[0021] FIG. 7 shows a flowchart of an interaction method according to some embodiments of the present disclosure;

[0022] FIG. 8 shows a flowchart of an interaction method according to some embodiments of the present disclosure;

[0023] FIG. 9 shows a block diagram of an interaction apparatus according to some embodiments of the present disclosure;

[0024] FIG. 10 shows a block diagram of an interaction apparatus according to some embodiments of the present disclosure;

[0025] FIG. 11 shows a block diagram of an electronic device according to some embodiments of the present disclosure.

[0026] It should be understood that the dimensions of the various portions shown in the drawings are shown for purposes of convenience and do not necessarily correspond to actual proportions. Wherever possible, the same reference numbers will be used throughout the drawings to depict the same or similar components. Thus, when a component is first introduced in a drawing, that component can not be discussed further in subsequent drawings. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only some of the embodiments of the present disclosure, but not all the embodiments. The description of the embodiments below is actually only illustrative, and should not be construed as any limitation on the present disclosure and its application or use. It should be understood that the present disclosure can be implemented in various forms, and should not be construed as being limited to the embodiments described herein.

[0028] It should be understood that various steps in the method implementations of the present disclosure can be performed in different order and / or in parallel. Additionally, the method implementations can include additional steps and / or omit performing the steps shown. The scope of the present disclosure is not limited in this regard. The relative arrangement of components and steps, numerical expressions, and numerical values set forth in these examples are to be interpreted as examples only and not limiting of the scope of the present disclosure unless otherwise specifically stated.

[0029] The term "include," and derivations thereof, means the term "comprise" or "contain" and variations as an open-ended term such that when the phrase "includes (comprises, contains)" is used, it means at least the stated elements, but not excluding others. Further, the term "comprise" and variations thereof as used in the present disclosure means the term "comprise" or "contain" and variations as an open-ended term such that when the phrase "comprises (contains)" is used, it means at least the stated elements, but not excluding others. Thus, include and comprise are synonymous. The term "based on" means "based, at least in part, on."

[0030] Reference throughout this specification to "an embodiment", "some embodiments" or "embodiments" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the application. The appearances of the phrases "in one embodiment" or "in some embodiments" or "in embodiments" in various places in the specification are not necessarily all referring to the same embodiment, although they can. Furthermore, the terms "a" or "an", as used herein, mean "one or more" or "at least one".

[0031] It should be noted that the terms "first", "second", and so on used in the present disclosure are merely used to distinguish different apparatuses, modules or units, and do not imply the order or the mutual dependency of the functions performed by these apparatuses, modules or units. Unless otherwise specified, the terms "first", "second", and so on are not intended to imply a given order or any other way of given order.

[0032] It should be noted that the terms "one", "multiple" used in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that "one" or "multiple" should be understood as "one or more" unless otherwise explicitly specified in the context.

[0033] The names of the messages or information exchanged between the plurality of apparatuses in the embodiments of the present disclosure are merely used for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0034] The embodiments of the present disclosure are described in detail below with reference to the drawings, but the present disclosure is not limited to these specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments. In addition, in one or more embodiments, specific features, structures, or characteristics can be combined by any suitable means from the present disclosure that will be clear to those of ordinary skill in the art.

[0035] As shown in FIG. 1, the interaction method includes: step S1, in response to receiving an input question of a user, searching according to the input question to obtain a search result; step S2, determining a search intent complexity level of the input question, wherein different search intent complexity levels correspond to different first machine learning models; step S3, processing the search result based on the first machine learning model corresponding to the search intent complexity level of the input question to generate an answer to the input question, wherein different search intent complexity levels correspond to different processing manners.

[0036] When receiving the input question of the user, the input of the user can be received through a keyboard, a graphical interface, etc., or the voice input of the user can be detected, etc., and the embodiments of the present disclosure do not limit this.

[0037] The searching according to the input question is, for example, searching through a search system, a search engine, etc.

[0038] The search intent complexity level of the input question is, for example, the complexity of the search intent implied by the input question of the user. Different input questions have different user needs, and the complexity of searching and answering also differs.

[0039] The first machine learning model is, for example, a generative large language model, which is used to understand the question of the user and generate an answer. For questions of different search intent complexity levels, different first machine learning models are used for processing, for example, to improve the adaptation between the question and the model, so that the machine learning model can better understand and answer the question, thereby improving the accuracy of the answer.

[0040] At the same time, for questions of different search intent complexity levels, different processing manners are used for the search results. That is, according to the complexity level of the question, different answering ideas are used, thereby improving the efficiency and accuracy of the answer.

[0041] According to the embodiments of the present disclosure, the search intent complexity level of the input question of the user is determined, and for the search results of the input questions of different search intent complexity levels, different processing manners are used based on different first machine learning models to generate answers, thereby improving the efficiency and accuracy of the answers.

[0042] The interaction method of the embodiments of the present disclosure can be executed on a server, or partially on a client.

[0043] The interactive method according to some other embodiments of the disclosure will be described below in combination with FIG. 2.

[0044] As shown in FIG. 2, in response to determining that the input question is of the first search intent complexity level in step S2, a new content is generated as an answer to the input question based on the first machine learning model corresponding to the first search intent complexity level according to the search result in step S31.

[0045] The search intent complexity levels include, for example, the first search intent complexity level and the second search intent complexity level. The first search intent complexity level is higher than the second search intent complexity level, that is, the question of the first search intent complexity level is more complex than the question of the second search intent complexity level.

[0046] It should be clear to those skilled in the art that the first search intent complexity level and the second search intent complexity level herein are only examples, and the questions can also be divided into three or more search intent complexity levels, for example, a third search intent complexity level, etc.

[0047] The case where the input question is of the first search intent complexity level will be described first.

[0048] The processing mode corresponding to the first search intent complexity level is more complex, for example, calculation, reasoning, comparison, prediction, evaluation, creation, etc. The processing mode simulates the complex thinking process of human beings. In step S31, the new content in the answer indicates that the new content is included relative to the search result. Compared with the processing mode such as abbreviation, modification of grammatical errors, summary and induction, etc., the calculation, reasoning, comparison, prediction, evaluation, creation, etc. processing mode corresponding to the first search intent complexity level not only integrates the existing search result, but also creates new content as the answer.

[0049] At the same time, the first machine learning model corresponding to the first search intent complexity level is also a model that is trained to be more suitable for answering difficult questions.

[0050] By first diverting the question, the complex question is diverted to the complex processing link, and the more complex processing mode is used, so that the adaptation degree of the processing mode and the generation model to the user's intent is higher, and the accuracy of the answer is improved. At the same time, by using the processing mode corresponding to the first search intent complexity level, new content is created, which can assist the user in decision-making and also provide new inspiration and materials for the user.

[0051] In some embodiments, in response to receiving the input question of the user, and the input question is of the first search intent complexity level, one or more searches are performed according to the input question to obtain the search result.

[0052] For example, the input question with different search intention complexity levels corresponds to different search manners. The search intention complexity level of the input question can be determined first, and then the search manner is determined according to the search intention complexity level of the input question.

[0053] When the input question is a relatively complex question, the result obtained by one or more searches is taken as the search result. Compared with single search, multiple search can provide more comprehensive and extensive information for answering difficult questions, thereby reducing the difficulty of answering and improving the accuracy of answering.

[0054] When searching according to the input question, the input question can be directly used for searching. The machine learning model can also be used to generate search words for the input question, and the search words can be used for searching.

[0055] FIG. 3 shows a schematic diagram of searching according to an input question according to some embodiments of the present disclosure.

[0056] The search manner when the input question is the first search intention complexity level will be described below in combination with FIG. 3.

[0057] As shown in FIG. 3, step S1 includes: step S11, in response to receiving the input question of the user and the input question being the first search intention complexity level, generating search words according to the input question based on the second machine learning model corresponding to the first search intention complexity level; step S12, searching by using the search words to obtain initial search results; step S13, determining whether it is necessary to search again according to the input question and the initial search results based on the third machine learning model corresponding to the first search intention complexity level; and step S14, in response to the need for searching again, determining the initial search results obtained by multiple searches as the search results.

[0058] The second machine learning model is, for example, a generation model. After the user inputs the question, search words are generated according to the input question based on the second machine learning model corresponding to the first search intention complexity level according to the input question. The search engine is used to search according to the search words to obtain the initial search results of this time.

[0059] The second machine learning models corresponding to different search intention complexity levels can be different. For example, the search words generated by the second machine learning model corresponding to the first search intention complexity level can be words that are not in the original question. The search words generated by the second machine learning model corresponding to the second search intention complexity level can be words in the original question.

[0060] Sometimes, the interaction with the user is a multi-turn dialogue, and in this case, the context information of the multi-turn dialogue (e.g., the question and answer of the previous turns) and the question of the current turn are jointly input into the second machine learning model corresponding to the first search intent complexity level when generating the search term. The second machine learning model understands and integrates the context information and the question of the current turn, and outputs one or more search terms.

[0061] Instead of using the second machine learning model to generate the search term, a regular, predetermined strategy, or the like can also be used. The input question can also be directly used to search in the search engine.

[0062] If the input question is of the first search intent complexity level, it is determined whether multiple searches are to be performed.

[0063] The third machine learning model is, for example, a discriminative model. For a question of the first search intent complexity level, it is determined whether multiple searches are to be performed based on the third machine learning model.

[0064] If it is determined that multiple searches are to be performed, multiple searches are performed, and the initial search results accumulated by the multiple searches are determined as the search results. Through serial multiple searches, more comprehensive and in-depth information is searched, thereby providing the user with more comprehensive and systematic answers.

[0065] If it is determined that multiple searches are not to be performed, the current search results are determined as the search results.

[0066] Meanwhile, for a question of the first search intent complexity level, sometimes serial multiple searches are needed, and the content of the final search results is relatively large, so more information of the first machine learning model needs to be input when generating the answer. Therefore, the first machine learning model corresponding to the first search intent complexity level can have a larger context window than the first machine learning model corresponding to the second search intent complexity level.

[0067] In some embodiments, in response to the need for further searching, the initial search results obtained by the multiple searches are determined as the search results, including: in response to the need for further searching, generating a new search term based on the previous initial search results and the input question according to the second machine learning model corresponding to the first search intent complexity level; performing a search using the new search term to obtain the current initial search results; determining whether further searching is needed based on the third machine learning model corresponding to the first search intent complexity level according to the previous initial search results and the current initial search results; in response to the need for further searching, repeating the generation of the new search term, the search using the new search term, and the determination of whether further searching is needed until further searching is not needed, and the initial search results obtained by the multiple searches are determined as the search results.

[0068] For example, the third machine learning model determines whether to continue searching according to the current initial search result and all previous initial search results.

[0069] If it is determined to continue searching, the step of generating a new search term is returned, and searching is performed using the new search term. Then, it is determined whether to search again according to the previous search result and the current initial search result. This cycle is repeated until it is determined that no further searching is required. All the initial search results obtained previously are determined as search results to be used subsequently.

[0070] Until it is determined that no further searching is required, the first machine learning model corresponding to the first search intent complexity level is used to generate an answer according to all the search results.

[0071] In some embodiments, in response to the need for further searching, a new search term is generated according to the previous initial search result, the input question, and the second machine learning model corresponding to the first search intent complexity level. Searching is performed using the new search term to obtain the current initial search result. An initial answer is generated according to the input question, the current initial search result, and the previous initial search result based on the first machine learning model corresponding to the first search intent complexity level. It is determined whether further searching is required based on the third machine learning model corresponding to the first search intent complexity level according to the initial answer generated this time. In response to the need for further searching, the generation of a new search term, the use of the new search term for searching, the generation of an initial answer, and the determination of whether further searching is required are repeated until no further searching is required. The initial search results obtained through multiple searches are determined as search results.

[0072] For example, after obtaining the current search result, the first machine learning model corresponding to the first search intent complexity level is called to generate an initial answer.

[0073] The third machine learning model determines whether to continue searching according to the initial answer generated this time. The third machine learning model determines whether to continue searching, for example, according to whether the initial answer generated this time meets a predetermined target.

[0074] If it is determined to continue searching, the step of generating a new search term is returned, and searching is performed using the new search term. Then, it is determined whether to search again according to the previous search result and the current initial search result. This cycle is repeated until it is determined that no further searching is required. All the initial search results obtained previously are determined as search results to be used subsequently.

[0075] Until it is determined that the search is not needed again, the first machine learning model corresponding to the first search intent complexity level is reused to generate an answer according to all the search results, as an answer to the input question of the user.

[0076] The interactive method according to some other embodiments of the present disclosure will be described below in combination with FIG. 4.

[0077] The case where the input question is of a second search intent complexity level will be described below, where the first search intent complexity level is higher than the second search intent complexity level.

[0078] As shown in FIG. 4, in response to determining in step S2 that the input question is of the second search intent complexity level, in step S32, the partial content of the search results is extracted based on the first machine learning model corresponding to the second search intent complexity level to obtain an answer to the input question.

[0079] Compared with the first search intent complexity level, the processing manner corresponding to the second search intent complexity level is simpler, for example, the first machine learning model performs abbreviation, modification of grammatical errors, summarization and induction on the search results in order to optimize the user reading experience and adapt to the reply scene.

[0080] The question is first diverted, and simple questions are diverted to a simple processing link, in which a relatively simple answer idea is used to generate an answer, which can improve processing efficiency and reduce resource consumption. At the same time, compared with using complex thinking to solve simple problems, using simple thinking to answer simple questions can reduce misunderstanding and redundant steps, and reduce the risk of additional complexity and errors.

[0081] In some embodiments, in response to receiving the input question of the user, searching according to the input question to obtain search results includes: in response to receiving the input question of the user, and the input question being of the second search intent complexity level, performing a single search according to the input question to obtain an answer to the input question.

[0082] For example, the search intent complexity level of the input question can be determined first, and then according to the search intent complexity level of the input question, it is determined which search method is used.

[0083] Not all questions rely on a serial search structure and a large amount of content input. Through the preposed search intent diversion, simple links can be used to solve simple general problems, and complex links can be used to solve complex problems. For simple questions, the search can be performed only once to obtain search results and generate an answer. Compared with complex problems, for simple problems, the step of judging whether to search multiple times can be omitted, thereby improving the answer efficiency for simple questions and reducing information interference and redundancy.

[0084] In some embodiments, in response to receiving the input question of the user, performing a single search according to the input question to obtain an answer to the input question includes: in response to receiving the input question of the user, and the input question being of a second search intent complexity level, generating a search term according to the input question based on a second machine learning model corresponding to the second search intent complexity level, wherein the second machine learning model corresponding to different search intent complexity levels is different; and performing a single search using the search term to obtain a search result.

[0085] For example, based on the second machine learning model corresponding to the second search intent complexity level, a keyword is extracted from the user question as a search term. Then, the search engine is used to search the keyword, and the result of this search is determined as the search result. The step of determining whether to search multiple times is omitted, and the first machine learning model directly generates an answer according to the result of this search when generating the answer.

[0086] In some embodiments, the search intent complexity level of the input question is determined according to at least one of whether the input question has a fixed answer or no fixed answer, whether the input question belongs to a specific field, the number of searches required to answer the input question, and whether the input question requires new content to be generated according to the search result.

[0087] For example, the process of processing the user input question is divided into two parts: demand understanding and information query, and information integration. The model first understands the question, disassembles the question, and then starts information query. New information is continuously added during the query process, so as to iteratively correct the understanding of the initial question and the expectation of the reply. Through processing of the information obtained by query, the final answer is obtained.

[0088] For the model, "demand understanding and information query" can be understood as "input", and "information processing" can be understood as "output". There are some important judgment factors in "input" and "output" for distinguishing different complexity levels of questions.

[0089] The judgment factors of "input" include, for example, whether the user's demand itself is easy to understand (for example, whether the demand is clear, whether it depends on prior knowledge), and whether the demand needs to be implemented through multiple searches.

[0090] The judgment factors of "output" include, for example, whether new content needs to be generated according to the search result. The judgment factors of "output" also include, for example, whether the question has convergent, uniquely determined objective answers or open questions.

[0091] In some embodiments, the fourth machine learning model is used to determine the search intent complexity level of the input question.

[0092] For example, the trained fourth machine learning model can be used to determine the search intent complexity level of the input question. The fourth machine learning model is, for example, a classification model.

[0093] In some embodiments, the search intent complexity level of the input question is determined based on preset rules.

[0094] For example, the search intent complexity level of the input question is determined by preset some discrimination strategies and rules.

[0095] As shown in FIG. 5, in terms of “input”, if the answer to the question depends on prior knowledge, for example, the question is a question in a certain professional field, and the knowledge in the professional field is needed, then the question is classified as complex. If the answer to the question does not depend on prior knowledge, but needs multiple online searches, then it is classified as medium. If the answer to the question does not depend on prior knowledge and does not need multiple online searches, then it is classified as simple.

[0096] In terms of “input”, questions such as “What will the weather be like tomorrow?”, “What is the English voiceover of this character in the game?”, “What is the origin of Xiaogong Chan Temple in Duofu Mountain?”, “How many Chinese yuan is 380 US dollars?”, “What are the highlights of this book?” and the like are simple questions. “What is the customer service phone number of the airline?”, “In terms of user base, rank these social software”, “What is the growth rate of the number of aircraft of these airlines compared to last year?” “How much is the stock price difference between these two companies?”, “What are the popular colors this year?”, “How to choose the color of a coat?”, “Recommend a camera suitable for beginners” are medium questions. “How to solve this problem in this professional field?”, “What is the interest on 100,000 yuan for one year?”, “Does social security stop after 15 years of payment?”, “Which stock of these two companies is better to buy?”, “Predict the impact of the release of this model on the existing video market” are complex questions.

[0097] In terms of “output”, if it is necessary to generate new content (i.e., integration) based on the search results, then it is determined whether the question is a convergent question. If integration is needed and it is a convergent question, then it is classified as medium. If integration is needed and it is an open question, then it is classified as complex. If no integration based on the search results is needed, then it is classified as simple.

[0098] If a question is classified as medium or complex in terms of input, and also classified as medium or complex in terms of output, then the question is of the first search intent complexity level, for example, the question circled by the dashed line in FIG. 5 is of the first search intent complexity level. If a question is classified as simple in terms of input and / or output, then the question is of the second search intent complexity level.

[0099] In some embodiments, the interaction method further comprises: in response to receiving a display request of the search result from the user, displaying the search result and / or the source of the search result to the user. This will be described below in connection with Figs. 6A-6B.

[0100] As shown in Fig. 6A, in the display interface, not only the question and the answer are displayed, but also a display button of the search result is displayed. If the user clicks the button, the display interface of Fig. 6B is entered. In the display interface of Fig. 6B, the search result and / or the source of the search result is expanded for the user to refer. For example, the source of the search result is displayed in the form of a link, and the user can select the corresponding link to jump to the webpage of the source of the search result.

[0101] In some embodiments, in response to receiving the input question from the user, searching according to the input question to obtain the search result comprises: in response to receiving the input question from the user, determining whether the input question needs to be searched through the network based on a fifth machine learning model. In response to determining that the input question needs to be searched through the network, searching according to the input question to obtain the search result.

[0102] The fifth machine learning model is, for example, a discriminative model. Before the foregoing steps S1-S3, it can be determined whether the input question needs to be searched through the network based on the fifth machine learning model. If the input question needs to be searched through the network, the steps S1-S3 are entered. The question that needs to be searched through the network is, for example, a question that is difficult to answer according to historical data, such as “What is the weather tomorrow?”.

[0103] In some embodiments, in response to determining that the input question does not need to be searched through the network, generating an answer to the input question based on a sixth machine learning model.

[0104] For example, the sixth machine learning model is a local generative model. If the input question does not need to be searched through the network, the local generative model is directly used to generate an answer to the question.

[0105] The question that needs to be searched through the network is, for example, a question that can be answered by the local model according to the saved historical knowledge or historical experience, such as “How many seasons are there in a year?”.

[0106] By pre-classifying the questions according to whether they need to be searched through the network, the questions that need to be searched through the network are processed through the network, and the questions that do not need to be searched through the network are processed locally, which can further improve the processing efficiency.

[0107] The training method of each machine learning model will be described below taking the search intent complexity level including the first search intent complexity level and the second search intent complexity level as an example.

[0108] First, the training method of the first machine learning model corresponding to different search intent complexity levels will be described.

[0109] The first training sample includes an input question and an answer of the first search intent complexity level. The first machine learning model corresponding to the first search intent complexity level is trained by using the first training sample. The second training sample includes an input question and an answer of the second search intent complexity level. The first machine learning model corresponding to the second search intent complexity level is trained by using the second training sample. The input question is used as an input feature, and the answer is used as a label.

[0110] The first training sample is a more complex question than the second training sample. The first machine learning model for solving complex questions is trained by using complex questions, so that the first machine learning model for solving complex questions can learn to understand the nature of the question more deeply and learn to provide more accurate answers through more and more complex steps.

[0111] The first machine learning model for solving simple questions is trained by using simple questions, so that the first machine learning model for solving simple questions can directly hit the point through fewer steps and provide more accurate answers to the user more quickly.

[0112] If the search result is obtained by searching through the second machine learning model to generate the keyword, the features of the first training sample and the second training sample can further include the search result corresponding to the question of each sample.

[0113] Similar to the first machine learning model, different search intent complexity levels also correspond to different second machine learning models. The second machine learning model corresponding to the first search intent complexity level is trained by using the question and the search keyword corresponding to the first search intent complexity level. The second machine learning model corresponding to the second search intent complexity level is trained by using the question and the search keyword corresponding to the second search intent complexity level. The question is used as an input feature, and the search keyword is used as a label.

[0114] The third machine learning model is trained by using the question, the answer, and whether the answer meets the target corresponding to the first search intent complexity level. The question and the answer are used as input features, and whether the answer meets the target is used as a label.

[0115] The fourth machine learning model is trained by using the question as an input feature and the search intent complexity level of the question as a label.

[0116] In the training process of the first to fourth machine learning models, the questions used as training data are, for example, questions that need to be answered online.

[0117] The question whether the question needs to be answered online can also be classified. The fifth machine learning model is trained by using the question as an input feature and whether the question needs to be answered online as a label.

[0118] The sixth machine learning model is trained with the question without networking requirement as an input feature and the answer to the question as a label.

[0119] The following takes the search intent complexity level including the first search intent complexity level and the second search intent complexity level as an example, and introduces the interaction method according to some embodiments of the present disclosure in combination with FIG. 7.

[0120] The question input by the user and the context of the interaction are taken as prompt information, and it is determined according to the prompt information whether the input question has a networking intent. If there is no networking intent, the local sixth machine learning model is used to generate an answer. If there is a networking intent, the question is searched in the search engine to obtain a search result. The search intent complexity level of the input question is determined. Questions of the first search intent complexity level and questions of the second search intent complexity level are processed in different links respectively.

[0121] If it is a question of the first search intent complexity level, it enters the link corresponding to the first search intent complexity level. In the link corresponding to the first search intent complexity level, the search result is integrated based on the first machine learning model corresponding to the first search intent complexity level to generate new content, and an answer is obtained.

[0122] If it is a question of the second search intent complexity level, it enters the link corresponding to the second search intent complexity level. In the link corresponding to the second search intent complexity level, part of the content of the search result is extracted based on the first machine learning model corresponding to the second search intent complexity level to obtain an answer.

[0123] The following introduces the interaction method according to some other embodiments of the present disclosure in combination with FIG. 8.

[0124] The question input by the user and the context of the interaction are taken as prompt information, and it is determined according to the prompt information whether the input question has a networking intent. If there is no networking intent, the local sixth machine learning model is used to generate an answer. If there is a networking intent, the search intent complexity level of the question is determined, and questions of the first search intent complexity level and questions of the second search intent complexity level are processed in different links respectively.

[0125] If the input question is of the first search intent complexity level, the first search intent complexity level corresponding link is entered. A search word is generated by using the second machine learning model corresponding to the first search intent complexity level, and the search engine is searched by using the search word. According to the search result, whether to continue searching is determined based on the third machine learning model. If it is needed to continue searching, the step of generating the search word by using the second machine learning model corresponding to the first search intent complexity level is returned. If it is not needed to continue searching, a new content is generated as an answer based on the first machine learning model corresponding to the first search intent complexity level.

[0126] If the input question is of the second search intent complexity level, the second search intent complexity level corresponding link is entered. A search word is generated by using the second machine learning model corresponding to the second search intent complexity level, and the search engine is searched by using the search word. According to the search result, a part of the search result is extracted to generate an answer based on the first machine learning model corresponding to the second search intent complexity level.

[0127] Some embodiments of the present disclosure provide an interaction device, which is described below in combination with FIG. 9.

[0128] As shown in FIG. 9, the interaction device 9 includes: a search module 91 configured to, in response to receiving an input question of a user, search according to the input question to obtain a search result, for example, for performing the step S1 of FIG. 1; a determination module 92 configured to determine a search intent complexity level of the input question, wherein different search intent complexity levels correspond to different first machine learning models, for example, for performing the step S2 of FIG. 1; and a generation module 93 configured to process the search result based on the first machine learning model corresponding to the search intent complexity level of the input question to generate an answer to the input question, wherein different search intent complexity levels correspond to different processing manners, for example, for performing the step S3 of FIG. 1.

[0129] In some embodiments, the interaction device further includes a display module configured to, in response to receiving a display request of the search result of the user, display the search result and / or a source of the search result to the user.

[0130] In some embodiments, the generation module is configured to, in response to the answer to the input question not needing to search through a network, generate the answer to the input question based on a sixth machine learning model.

[0131] An interaction device according to some other embodiments of the present disclosure is described below in combination with FIG. 10.

[0132] As shown in FIG. 10, the interaction device 10 comprises a memory 101, and a processor 102 coupled to the memory 101, the processor 102 is configured to execute the interaction method of any of the preceding embodiments based on instructions stored in the memory 101.

[0133] The memory 101 is configured to store one or more computer-readable instructions. The memory 101 can comprise any combination of various types of computer-readable storage media, such as volatile memory and / or non-volatile memory, including but not limited to random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), read-only memory (ROM), flash memory. The memory 101 can store, for example, an operating system, an application program, a boot loader, a database, and other programs, and can also store various application programs and various data.

[0134] The processor 102 is configured to run the computer-readable instructions to implement the interaction method of any of the preceding embodiments. The specific implementation of each step of the interaction method can refer to the above-mentioned embodiments, and the repeated parts will not be described here.

[0135] The processor 102 can be configured to execute steps S1-S4 of FIG. 1 or steps S1'-S3' of FIG. 10. The processor 102 can be embodied as various processing devices, such as a central processing unit (CPU), a network processor (NP), etc.; and can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The central processing unit (CPU) can be X86 or ARM architecture, etc.

[0136] The processor 102 and the memory 101 can communicate with each other directly or indirectly. For example, the processor 102 and the memory 101 can communicate through a network. The network can include a wireless network, a wired network, and / or any combination of a wireless network and a wired network. The processor 102 and the memory 101 can also communicate with each other through a system bus, and the present disclosure does not limit the communication between the processor 102 and the memory 101.

[0137] It should be noted that the components of the interaction device 10 shown in FIG. 10 are only exemplary and not limiting, and the interaction device 10 can also have other components according to actual application needs. The processor 102 can control other components in the interaction device 10 to perform the desired functions.

[0138] The interaction device can be implemented by software, firmware and / or hardware, and can be integrated into an electronic device installed with a related application program.

[0139] An electronic device according to some embodiments of the disclosure will be described below with reference to FIG. 11.

[0140] The electronic device 11 shown in FIG. 11 can be a computer system having a dedicated hardware structure, and can perform a corresponding function when a related application is installed.

[0141] The electronic device includes, but is not limited to, a mobile terminal such as a smartphone, a notebook computer, a Personal Digital Assistant (PDA), a Tablet Personal Computer (Tablet PC), a PMP (Portable Multimedia Player), a car terminal (e.g., a car navigation terminal), a wearable device, and the like, and a stationary terminal such as a digital television, a desktop computer, and the like.

[0142] As shown in FIG. 11, a central processing unit (CPU) 111 performs various processes according to a program stored in a read-only memory (ROM) 112 or a program loaded from a storage portion 118 to a random access memory (RAM) 113. In the RAM 113, data required when the CPU 111 performs various processes and the like is stored as necessary. The central processing unit is merely exemplary, and can be other types of processors such as various processors described above. The ROM 112, the RAM 113, and the storage portion 118 can be various forms of computer-readable storage media. Note that although the ROM 112, the RAM 113, and the storage portion 118 are shown separately in FIG. 11, one or more of them can be combined or located in the same or different memory or storage module.

[0143] The CPU 111, the ROM 112, and the RAM 113 are connected to each other via a bus 114. An input / output interface 115 is also connected to the bus 114.

[0144] The following components are connected to the input / output interface 115: an input portion 116 including a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, and the like; an output portion 117 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), a speaker, a vibrator, and the like; a storage portion 118 including a hard disk, a magnetic tape, and the like; and a communication portion 119 including a network interface card such as a LAN card, a modem, and the like. The communication portion 119 allows communication processing to be performed via a network such as the Internet. It is readily understood that although the various devices or modules in the electronic device 11 are shown in FIG. 11 as communicating via the bus 114, they can also communicate through a network or other means, where the network can include a wireless network, a wired network, and / or any combination of a wireless network and a wired network.

[0145] The drive 1110 is also connected to the input / output interface 115 as necessary. A removable recording medium 1111 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is installed in the drive 1110 as necessary, so that a computer program read out from the removable recording medium 1111 is installed in the storage portion 118 as necessary.

[0146] In the case where the above series of processes are implemented by software, the program constituting the software can be installed from a network such as the Internet or a storage medium 1111 such as the removable recording medium 1111.

[0147] According to an embodiment of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product that, when executed on a computer, causes the computer to implement the interaction method described in any of the preceding embodiments. The computer program product includes a computer program carried on a computer-readable medium, which contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 119, or installed from the storage portion 118, or installed from the ROM 112. When the computer program is executed by the CPU 111, the interaction method of the embodiment of the present disclosure is executed.

[0148] Note that, in the context of the present disclosure, the computer-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device.

[0149] The computer-readable medium can be a computer-readable storage medium, or a computer-readable signal medium, or any combination of the two.

[0150] The computer-readable storage medium includes, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination of the foregoing. More specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium stores a computer program that, when executed by a processor, implements the interaction method described in any of the preceding embodiments.

[0151] The computer readable medium can include a computer-readable signal medium and / or a computer-readable storage medium. A computer readable signal medium can include a propagated data signal with computer executable code with a baseband component, with variations, or with a carrier component. Computer readable storage media can include, but are not limited to, volatile memory, non-volatile memory, magnetic media, optical media, and / or other storage media. Common forms of computer-readable media include, for example, a floppy disk, a flexible disk, hard disk, magnetic tape, or magnetic cassette, an optical disk, a compact disk, a compact disk-read only memory (CD-ROM), and the like, a shrewd medium, a memory card, a digital video disk (DVD), a Blu-ray disk, the Internet or other network distribution, any appropriate medium, or any suitable combination of the foregoing. The computer program product includes a computer readable medium.

[0152] The computer readable medium can be included within the electronic device; or can exist separately from the electronic device.

[0153] In some embodiments, a computer program including instructions which, when executed by a processor, causes the processor to carry out the interaction method of any one of the above embodiments is also provided. For example, the instructions can be embodied in computer program code.

[0154] In some embodiments, a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the interaction method according to any one of the above embodiments is also provided.

[0155] In some embodiments, a computer program product including computer program instructions which, when executed by a processor, implement the interaction method according to any one of the above embodiments is also provided. For example, the instructions can be embodied in computer program code.

[0156] In embodiments of the present disclosure, the computer program code for carrying out operations of the present disclosure can be written in one or more programming languages or combinations of languages including object oriented, scripting and / or procedural languages such as Java, C++, Smalltalk, Python, VB.NET, or similar language. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). The computer program code can also be embodied in a computer readable storage medium that can be any media that can be read and used by a computer.

[0157] The computer program product of the first aspect can include one or more non-transitory computer-readable media storing instructions that, when executed by one or more processors of a computing device, cause the one or more processors to perform the operations of the method of the first aspect. The one or more non-transitory computer-readable media can include one or more of the following: a magnetic storage device, an optical storage device, a solid-state storage device, a hard disk drive, a flash drive, a RAM, a ROM, a database, and a cache. The one or more processors can include one or more of the following: a central processing unit (CPU), a microprocessor, a microcontroller, a microcomputer, a microprocessor-based or microcontroller-based system, a programmable logic unit (PLU), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a system-on-chip (SoC), and a complex programmable logic device (CPLD).

[0158] The functions described above can be performed, at least in part, by one or more hardware logic components. For example, and without limitation, non-limiting examples of hardware logic components that can be used include field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SOCs), complex programmable logic devices (CPLDs), etc.

[0159] While certain aspects of the present disclosure have been described with reference to particular examples, those having ordinary skill in the art will appreciate that the examples are illustrative of the disclosure and are not meant to limit the scope of the disclosure. Various modifications to the examples described in this disclosure can be made by persons with ordinary skill in the art without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. An interaction method, comprising: in response to receiving an input question of a user, searching according to the input question to obtain a search result; determining a search intention complexity level of the input question, wherein different search intention complexity levels correspond to different first machine learning models; processing the search result based on the first machine learning model corresponding to the search intention complexity level of the input question to generate an answer to the input question, wherein different search intention complexity levels correspond to different processing manners.

2. The interaction method of claim 1, wherein, The search intention complexity level includes a first search intention complexity level, and the processing the search result based on the first machine learning model corresponding to the search intention complexity level of the input question to generate an answer to the input question comprises: in response to the input question being the first search intention complexity level, generating new content as the answer to the input question according to the search result based on the first machine learning model corresponding to the first search intention complexity level.

3. The interaction method of claim 2, wherein, The searching according to the input question to obtain a search result in response to receiving an input question of a user comprises: in response to receiving an input question of a user and the input question being the first search intention complexity level, searching one or more times according to the input question to obtain the search result.

4. The interaction method of claim 3, wherein, The searching one or more times according to the input question to obtain a search result in response to receiving an input question of a user comprises: in response to receiving an input question of a user and the input question being the first search intention complexity level, generating a search word according to the input question based on a second machine learning model corresponding to the first search intention complexity level; searching using the search word to obtain an initial search result; determining whether to search again according to the input question and the initial search result based on a third machine learning model corresponding to the first search intention complexity level; in response to needing to search again, determining the initial search result obtained through multiple searches as the search result.

5. The interaction method of claim 4, wherein, The determining the initial search result obtained through multiple searches as the search result in response to needing to search again comprises: in response to needing to search again, generating a new search word according to the previous initial search result and the input question based on the second machine learning model corresponding to the first search intention complexity level; searching using the new search word to obtain a current initial search result; determining whether to search again according to the previous initial search result and the current initial search result based on the third machine learning model corresponding to the first search intention complexity level; in response to needing to search again, repeatedly generating a new search word, searching using the new search word, and determining whether to search again until not needing to search again, and determining the initial search result obtained through multiple searches as the search result.

6. The interaction method of claim 4, wherein, The determining the initial search result obtained through multiple searches as the search result in response to needing to search again comprises: In response to a need for re-search, based on the previous initial search result and the input question, a new search term is generated based on a second machine learning model corresponding to the first search intent complexity level; Search is performed using the new search term to obtain a current initial search result; Based on a first machine learning model corresponding to the first search intent complexity level, an initial answer is generated based on the input question, the current initial search result and the previous initial search result; Based on a third machine learning model corresponding to the first search intent complexity level, it is determined whether re-search is needed based on the generated initial answer; In response to a need for re-search, the generation of a new search term, the search using the new search term, the generation of an initial answer and the determination of whether re-search is needed are repeated until no re-search is needed, and the initial search results obtained through multiple searches are determined as the search result.

7. The interaction method of claim 2, wherein, The search intent complexity level includes a second search intent complexity level, and the processing of the search result based on the first machine learning model corresponding to the search intent complexity level of the input question to generate an answer to the input question includes: In response to the input question being the second search intent complexity level, the first machine learning model corresponding to the second search intent complexity level is used to extract part of the content of the search result to obtain an answer to the input question, wherein the first search intent complexity level is higher than the second search intent complexity level. The search based on the input question in response to receiving the input question of the user includes:

8. The interaction method of claim 7, wherein, In response to receiving the input question of the user and the input question being the second search intent complexity level, a single search is performed based on the input question to obtain an answer to the input question. The single search based on the input question in response to receiving the input question of the user includes:

9. The interaction method of claim 8, wherein, In response to receiving the input question of the user and the input question being the second search intent complexity level, a search term is generated based on a second machine learning model corresponding to the second search intent complexity level based on the input question, wherein the second machine learning models corresponding to different search intent complexity levels are different; The single search is performed using the search term to obtain a search result. The determination of the search intent complexity level of the input question includes:

10. The interaction method of claim 1, wherein, The search intent complexity level of the input question is determined based on a fourth machine learning model; or The search intent complexity level of the input question is determined based on a preset rule. The determination of the search intent complexity level of the input question includes:

11. The interaction method of claim 1, wherein, The search intent complexity level of the input question is determined based on at least one of whether the input question has a fixed answer or does not have a fixed answer, whether the input question belongs to a specific field, the number of times the input question needs to be searched, and whether the input question needs to generate new content based on the search result.

12. The interaction method of claim 1, further comprising: ​ In response to receiving the display request of the search result by the user, the search result and / or a source of the search result is displayed to the user.

13. The interaction method of claim 1, wherein, The searching according to the input question by the user comprises: In response to receiving the input question by the user, determining whether searching through the network is needed to answer the input question based on a fifth machine learning model; In response to determining that searching through the network is needed to answer the input question, searching according to the input question to obtain a search result.

14. The interaction method of claim 13, further comprising: In response to determining that searching through the network is not needed to answer the input question, generating an answer to the input question based on a sixth machine learning model.

15. The interaction method of claim 1, wherein, The search intent complexity level comprises a first search intent complexity level and a second search intent complexity level, and the interaction method further comprises: Obtaining a first training sample and a second training sample, wherein the first training sample comprises an input question and an answer of the first search intent complexity level, and the second training sample comprises an input question and an answer of the second search intent complexity level; Training a first machine learning model corresponding to the first search intent complexity level by using the first training sample; Training a first machine learning model corresponding to the second search intent complexity level by using the second training sample.

16. An interaction apparatus, comprising: a searching module configured to search according to an input question by a user to obtain a search result in response to receiving the input question by the user; a determining module configured to determine a search intent complexity level of the input question, wherein different search intent complexity levels correspond to different first machine learning models; a generating module configured to process the search result based on a first machine learning model corresponding to the search intent complexity level of the input question to generate an answer to the input question, wherein different search intent complexity levels correspond to different processing manners.

17. An interaction apparatus, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute instructions stored in the memory, to perform the interaction method according to any one of claims 1 to 15.

18. A computer readable storage medium having computer program instructions stored thereon, the instructions being executed by a processor to implement the interaction method according to any one of claims 1 to 15.

19. A computer program product comprising computer program instructions, the computer program instructions being executed by a processor to implement the interaction method according to any one of claims 1 to 15.

Images