Method, device, electronic device and storage medium for determining interaction information

Abstract

The present disclosure provides a method, device, electronic device and storage medium for determining interactive information, and relates to the field of artificial intelligence technology, and in particular to the fields of large models, generative models, NLP, intelligent search, etc. The specific implementation scheme is: according to the query information and historical query information of the object, multiple question dimensions are determined, each question dimension includes a dimension name and multiple options; according to whether the semantic information of the multiple question dimensions and the semantic information of the query results associated with the query information are consistent, and the evaluation values ​​of the multiple question dimensions, a target question dimension is determined from the multiple question dimensions; and according to the dimension name and multiple options in the target question dimension, the interactive information is determined.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to the fields of large models, generative models, NLP, intelligent search, etc. More specifically, the present disclosure provides a method, device, electronic device, storage medium and computer program product for determining interaction information. Background Art

[0002] In complex search scenarios, the diversity and ambiguity of user expressions will cause search engines to be unable to accurately capture the true search intent, which in turn leads to a deviation between search results and users' actual needs. Summary of the invention

[0003] The present disclosure provides a method, an apparatus, an electronic device, a storage medium, and a computer program product for determining interaction information.

[0004] According to one aspect of the present disclosure, a method for determining interaction information is provided, including: determining multiple question dimensions based on query information and historical query information of an object, each question dimension including a dimension name and multiple options; determining a target question dimension from the multiple question dimensions based on whether semantic information of the multiple question dimensions is consistent with semantic information of query results associated with the query information, and evaluation values ​​of the multiple question dimensions; and determining interaction information based on the dimension name and multiple options in the target question dimension.

[0005] According to another aspect of the present disclosure, a device for determining interactive information is provided, comprising: a first determination module, a second determination module, and a third determination module. The first determination module is used to determine multiple question dimensions based on the query information and historical query information of the object, each question dimension including a dimension name and multiple options. The second determination module is used to determine a target question dimension from multiple question dimensions based on whether the semantic information of the multiple question dimensions is consistent with the semantic information of the query results associated with the query information, and the evaluation values ​​of the multiple question dimensions. The third determination module is used to determine the interactive information based on the dimension name and multiple options in the target question dimension.

[0006] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method provided by the present disclosure.

[0007] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method provided by the present disclosure.

[0008] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method provided by the present disclosure is implemented.

[0009] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0011] Figure 1 is a schematic diagram of an application scenario of a method and device for determining interaction information according to an embodiment of the present disclosure;

[0012] Figure 2 is a schematic flow chart of a method for determining interaction information according to an embodiment of the present disclosure;

[0013] Figure 3A is a schematic diagram of a method for determining interaction information according to an embodiment of the present disclosure;

[0014] Figure 3B is a schematic diagram of an interactive page according to an embodiment of the present disclosure;

[0015] Figure 4 is a schematic structural block diagram of an apparatus for determining interaction information according to an embodiment of the present disclosure; and

[0016] Figure 5 It is a structural block diagram of an electronic device used to implement the method for determining interaction information according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0017] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0018] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0019] In the technical solution of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0020] In complex search scenarios, the diversity and ambiguity of user expressions will cause search engines to be unable to accurately capture the true search intent, which in turn leads to a deviation between search results and users' actual needs.

[0021] For example, the query information "Traditional Chinese medicine for body conditioning" is a relatively complex and vague demand, which involves the user's personalized preferences. However, the search engine will return relatively broad query results related to keywords such as traditional Chinese medicine and body conditioning, which contain a lot of content that does not meet the user's needs and cannot meet the user's personalized needs well. Some search engines will recommend other possible query information to users through related searches, but it is still difficult to discover the user's real needs in this way, and it is easy to mix content that is irrelevant to the user's needs.

[0022] The disclosed embodiment provides a method for determining interactive information, which can generate question dimensions for query information by combining the prior expert knowledge of the large model and the a posteriori supplement in the user's historical query information through the method of thought chain. By asking questions to the user, the user is guided to provide more detailed information about the query target, thereby exploring the user's query needs. In this way, the search engine can provide query results that match the user's needs, improve the accuracy and relevance of the query results, and provide users with a more personalized and satisfactory search experience. In addition, since the user can actively participate in the query process instead of passively accepting the content recommended by the search engine, the user's initiative and participation are improved, thereby improving the user experience.

[0023] The technical solution provided by the present disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Figure 1 It is a schematic diagram of an application scenario of the method and device for determining interaction information according to an embodiment of the present disclosure.

[0025] It should be noted that Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0026] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0027] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptops, desktop computers, etc.

[0028] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as interactive information generated according to the query information input by the user, etc.) to the terminal device.

[0029] It should be noted that the method for determining the interaction information provided in the embodiment of the present disclosure can generally be performed by the server 105. Accordingly, the apparatus for determining the interaction information provided in the embodiment of the present disclosure can generally be set in the server 105. The method for determining the interaction information provided in the embodiment of the present disclosure can also be performed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the apparatus for determining the interaction information provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0030] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0031] Figure 2 is a schematic flowchart of a method for determining interaction information according to an embodiment of the present disclosure.

[0032] like Figure 2 As shown, the method 200 for determining interaction information may include operations S210 to S230.

[0033] In operation S210 , a plurality of question dimensions are determined according to the query information and the historical query information of the object, each of which includes a dimension name and a plurality of options.

[0034] For example, a subject may include a user who uses a search engine.

[0035] For example, the query information may be content that needs to be queried and input by the user, and the data type of the query information may be text, for example, the query information is "recommendations for good novels".

[0036] For example, historical query information may include query information within a predetermined period of time in the past, and the predetermined period of time may be half a month. In some embodiments, some content supplemented by the user may be used as historical query information. For example, after a user enters a query information in a search engine, he does not exit the search engine and continues to enter some other content, or selects query content provided by some pages in the search engine, and the content may be used as historical query information.

[0037] For example, dimension names may include "novel type", "novel length", etc., and options associated with "novel type" may include "suspense and mystery", "science fiction", etc., and options associated with "novel length" may include "novel", "novel", "short film", etc.

[0038] For example, keywords may be extracted from the query information and historical query information, and a mapping relationship between keywords and question dimensions may be preconfigured, and then the question dimensions may be determined based on the mapping relationship.

[0039] In operation S220 , a target question dimension is determined from among the multiple question dimensions according to whether the semantic information of the multiple question dimensions is consistent with the semantic information of the query result associated with the query information and the evaluation values ​​of the multiple question dimensions.

[0040] For example, the query result may represent the content obtained by searching the query information. The data types of the query result may include text, image, video, etc. For example, the query result includes the following content: "The recommended novels include the first novel, the novellas include the second novel, and the short stories include the third novel", then the query content is semantically consistent with the "novel length" in the question dimension.

[0041] For example, the evaluation value can represent the degree of influence of the question dimension on the query content, and can also represent the degree of influence of the question dimension on understanding the user's query needs. The evaluation value can be a specific value, or it can be an evaluation parameter that represents the degree of the evaluation value. The evaluation value corresponding to the question dimension can be pre-configured, or a large model can be used for evaluation.

[0042] For example, a question dimension that is semantically inconsistent with the query result and has a higher evaluation value can be determined as a target question dimension.

[0043] In operation S230 , interaction information is determined according to the dimension name and multiple options in the target question dimension.

[0044] For example, the dimension name and multiple options in the target question dimension can be used as interactive information. For another example, the dimension name in the target question dimension can be expressed in natural language, and the dimension name "novel type" can be converted into "What type of novel do you want to read?", and then the converted dimension information and options are used as interactive information, so that the interactive information is easier for users to understand.

[0045] The disclosed embodiment guides the user to provide more detailed information about the query target by asking questions to the user, thereby exploring the user's query needs. In this way, the search engine can provide query results that match the user's needs, improve the accuracy and relevance of the query results, and provide the user with a more personalized and satisfactory search experience. In addition, since the user can actively participate in the query process instead of passively accepting the content recommended by the search engine, the user's initiative and participation are improved, thereby improving the user's experience.

[0046] In actual applications, after obtaining the interactive information, there are many ways to apply the interactive information. In one application scenario, the interactive information can be displayed on the front-end page so that users can make selections based on the question dimensions to supplement the query content. In another application scenario, the interactive information can be used to construct training samples, and the training samples can be used to train other models, so that other trained models can be used to interact with users in the online application. It is understandable that the resources required for other models can be less than those required for the large model, and the other models can be SFT (Sparse Fine Tuning) models.

[0047] According to another embodiment of the present disclosure, the historical query information may be screened first, and the screening process may include: determining the candidate historical query information that meets the predetermined screening condition as the historical query information, wherein the predetermined screening condition may include at least one of the following: the intention of the candidate historical query information is consistent with the intention of the query information; Figure 1 The search frequency of the candidate historical query information meets the predetermined frequency condition, and the entity included in the candidate historical query information is consistent with the entity included in the query information.

[0048] For example, historical query information within a predetermined period of time in the past may be obtained first, and the historical query information may be used as candidate historical query information.

[0049] Next, it can be determined whether the candidate historical query information is consistent with the query information. For example, it can be determined whether the intent is consistent by binary classification, such as inputting the candidate historical query information and the query information into a pre-trained classification model, and the classification model determines whether the intent is consistent. For another example, it is also possible to perform intent recognition on the candidate historical query information and the query information respectively, and then determine whether the intents of the two are consistent.

[0050] Next, the frequency of the candidate historical query information can be counted, and then the candidate historical query information can be filtered according to the predetermined frequency condition. It can be understood that the frequency can reflect the user's preference. The predetermined frequency condition can include: the frequency of the candidate historical query information is greater than a threshold, and the threshold can be a value such as 5. The predetermined frequency condition can also include: the frequency ranking of the candidate historical query information is in the top predetermined position, such as the frequency is in the top 10.

[0051] Next, it may be determined whether the entities included in the filtered candidate historical query information are consistent with the entities included in the query information. The method of determining the entities may include named entity recognition, which is not limited in this embodiment.

[0052] For example, the query information and candidate historical query information are "How to repair a broken mobile phone of the first brand" and "How to repair a broken mobile phone of the second brand", and the meanings of the two are similar. Figure 1 The entities are consistent, but not identical.

[0053] Based on the predetermined screening conditions, this embodiment can screen historical query information that has a high correlation with the query information and is of greater interest to the user from candidate historical query information, thereby ensuring the accuracy of question dimensions subsequently determined based on the historical query information.

[0054] Figure 3A is a schematic diagram of a method for determining interaction information according to an embodiment of the present disclosure, Figure 3B It is a schematic diagram of an interactive page according to an embodiment of the present disclosure.

[0055] According to another embodiment of the present disclosure, the method for determining interaction information may include operations S310 to S340, wherein operation S340 is performed before operation S310.

[0056] In operation S340, it is determined whether a question needs to be asked.

[0057] In operation S310 , a plurality of question dimensions are determined.

[0058] In operation S320 , the target question dimension is filtered.

[0059] In operation S330, interaction information is determined.

[0060] In the above operation S340, determining whether a question needs to be asked is a pre-processing process, which is mainly used to determine whether there are complex requirements and the query information needs to be supplemented by asking questions. Whether a question needs to be asked can be determined by a pre-processing prompt information template. For example, the query information input by the user and the pre-processing prompt information template can be combined and then input into the big model, and the big model outputs the determination result. The pre-processing prompt information template can be pre-configured and stored in a predetermined storage area, and read from the storage area when it is needed.

[0061] For example, the pre-processing prompt information template may include a thought chain, which includes multiple steps, and the thought chain instructs the large model to reason according to the steps in the thought chain. For example, the multiple steps included in the thought chain include: The core of the analysis is to determine whether the query results need to consider the distinction of different situations or the personalized analysis of users, such as personalized preferences, user personal portraits, etc. If so, it is determined that the current query belongs to a complex demand and needs to be asked, otherwise it is determined that no question is needed.

[0062] For another example, the pre-processing prompt information template may include examples, which mainly illustrate which are personalized situations and which are distinguished by different situations by giving examples. For example, the examples in the pre-processing prompt information template are as follows: Taking the query information of "recommendations for good novels" as an example, answering this question needs to take into account personalized factors such as the user's reading habits and favorite literary genres. Taking the query information of "Beijing house prices" as an example, due to the large differences in prices of different regions and different types of real estate, the answer needs to involve specific situational factors such as region and property type. These questions are complex needs and there is room for questions. On the contrary, taking the query information of "one hour is equal to how many hours" as an example, this type of demand is clear and will not change due to the situation or personal specific circumstances, so it is not a complex demand and there is no need to ask questions.

[0063] This embodiment ensures the accuracy of questions by determining the need for asking questions, so that more user information can be obtained through asking questions in appropriate situations, and more accurate personalized answers can be provided.

[0064] In the above operation S310, in the process of determining multiple question dimensions, multiple question dimensions can be determined based on the query information and historical query information of the object. In addition, two perspectives, namely, a priori refined dimension analysis and a posteriori refined dimension mining, can be considered in the determination process. For example, a large model can be prompted by a prompt information template, and the contextual factors and personalized factors that affect the answer to the query information can be disassembled from the two perspectives of a priori and a posteriori, and the dimension set of the question can be summarized. Refinable options are generated in combination with the demand dimension to form an interactive user selection interface.

[0065] For example, for the query information "English course recommendation", we first conduct a priori analysis. Since different learners have different personalized needs for the content and form of the course, we summarize the detailed dimensions as follows: course content, teaching method. A posteriori, based on the historical query information "Online English teaching in Beijing" and "English courses suitable for primary school students" formed by this question, we parse out the supplementary information including: Beijing, online, and primary school students, and summarize the question dimensions as follows: learning population, teaching method, and region. In this way, the question dimensions are constructed including: {"dimension name": "Course content", "options": ["Daily", "Business", "Exam"]}, {"dimension name": "Teaching method", "options": ["Online", "Offline"]}, {"dimension name": "Learning population", "options": ["Primary school students", "Middle school students", "Adults"]}, {"dimension name": "Region", "options": ["Beijing", "Shanghai", "Guangzhou"]}.

[0066] In this embodiment, on the one hand, by using the prior knowledge of the large model, direct analysis is performed on the original demand, which can provide insight into and inspire the user's comprehensive demand direction. On the other hand, combined with the user's historical search behavior posteriori, the user's real and specific needs are extracted to enhance the confidence of the question content and the applicability of the question. In this way, detailed option content can be generated for each question dimension according to demand. In the search interface, an interactive user selection area can be formed to achieve the effect of guiding users to clarify complex needs.

[0067] In the above operation S320, during the process of screening the target question dimension, the target question dimension can be determined from the multiple question dimensions based on whether the semantic information of the multiple question dimensions is consistent with the semantic information of the query result associated with the query information, and the evaluation values ​​of the multiple question dimensions.

[0068] In the above operation S330, in the process of determining the interaction information, the interaction information may be determined according to the dimension name and multiple options in the target question dimension.

[0069] According to another embodiment of the present disclosure, the process of determining multiple question dimensions may include: combining a first prompt information template, query information, historical query information, and query results associated with the query information into first prompt information, and then inputting the first prompt information into a large model to obtain multiple question dimensions. The first prompt information template may be pre-configured and stored in a predetermined storage area, and read from the storage area when needed.

[0070] In one example, the first prompt information template includes a first chain of thought, which includes multiple steps, and the first chain of thought instructs the large model to reason according to the steps in the first chain of thought. For example, the multiple steps included in the first chain of thought include: determining the relationship type between the query information and the historical query information, determining whether there is a question dimension based on the relationship type, and generating question dimensions and options based on a predetermined refinement direction in the case of candidate question dimensions. In other examples, the first prompt information template may also indicate the identity and task of the large model. The first prompt information template may also indicate the format of the input information and output information of the large model. The first prompt information template may also include examples. In this embodiment, the first prompt information template includes the first chain of thought, and guiding the large model through the steps in the first chain of thought can enable the large model to accurately output the question dimension.

[0071] For example, in this embodiment, the query information is "movies suitable for two people to watch", and the historical query information includes the following multiple information: movies suitable for two people to watch before exercise, movies for one person to watch, movies suitable for two people to watch in the United States, and movies suitable for couples to watch. Based on the above query information and historical query information, the following question dimensions can be determined: {'dimension name': 'Movie type', 'options': ['comedy', 'action', 'romance', 'science fiction']}, {'dimension name': 'movie origin', 'options': ['United States', 'China', 'India', 'other countries']}, {'dimension name': 'specific factors of concern', 'options': ['richness of plot', 'visual effects', 'reputation and evaluation']}.

[0072] For example, in this embodiment, the query information is "top ten mobile games", and the historical query information includes the following multiple information: mobile game rankings, top ten martial arts mobile games, top ten mobile games, and top 10 mobile games. Based on the above query information and historical query information, the following question dimensions can be determined: {'dimension name': 'game type', 'options': ['martial arts', 'shooting', 'strategy', 'adventure']}, {'dimension name': 'ranking basis', 'options': ['download volume', 'active user volume', 'user evaluation']}.

[0073] For example, in this embodiment, the first prompt information template is as follows:

[0074] You are a search query analysis expert. You need to ask questions about the "current query" based on the "current query" and "historical query information", refine the user's intention, and guide the user to conduct multiple rounds of interactive searches. "Refinement" is defined as: a specific meaning.

[0075] The specific steps of disassembly thinking are as follows:

[0076] Step 1: Problem demand analysis: Analyze what kind of information the user wants to obtain through this question, or what kind of problem the user wants to solve. And determine whether the demand for this question is clear.

[0077] Step 2, historical query information analysis: Analyze "historical query information" and determine their relationship with the "current query". The relationship is mainly divided into four categories. The first category is the same semantics as the "current query" and has no information gain. The second category is completely irrelevant to the "current query". The third category is that the subject or demand content has changed. The fourth category is further refined based on the "current query". Based on the above analysis, determine whether the user's historical query information can provide effective refinable dimensions. If so, output True, otherwise output False.

[0078] Step 3: Determine whether the problem needs to be refined: If step 2 is not False, you can refer to the content of "historical query information" during the analysis process. When making specific judgments, consider whether the current problem needs to be refined from three directions: "supplement specific situation refinement", "refinement based on user personal situation" and "refinement based on demand level". If refinement is not required in all three directions, it means that the problem is specific enough and does not need to be refined further. Otherwise, it can be determined that the current problem needs to be further refined. Determine whether the current problem needs to be refined based on the problem requirements and "historical query information content".

[0079] Step 4: List the detailed dimensions: List the detailed dimensions that have a key impact on answering the current question from the three directions of "supplementing specific situational details", "refining based on the user's personal situation" and "refining by level of needs". The detailed dimensions in the "refining by level of needs" direction do not need to have optional content, just give the detailed dimensions. The detailed dimensions in the other two directions need to provide optional content.

[0080] Step 5: Remove duplicates from the existing refined dimensions of the current query result: Analyze whether the current query result contains the refined dimensions decomposed above. If so, delete the semantically duplicated dimensions in the subsequent analysis.

[0081] ##Input format: query{}, Answer{}, historical query information{}.

[0082] ##Output format:

[0083] {

[0084] "Problem Requirement Analysis": "Analyze the user's "current query"".

[0085] "Historical query information analysis": "Analyze historical query information behavior".

[0086] "Determine whether the problem needs to be refined": "Determine whether refinement is triggered from three aspects."

[0087] "List refinement dimensions": "List refinement dimensions and corresponding options from three directions".

[0088] "The current query results have been refined to deduplicate": "".

[0089] }

[0090] ## Example: AAA.

[0091] ##Based on the requirements and [examples], identify your thought process and results.

[0092] The first prompt information template is described above. It can be understood that the query in the first prompt information template represents query information and the answer represents query result.

[0093] According to another embodiment of the present disclosure, the process of screening the target question dimensions may include: a deduplication sub-process and a value assessment sub-process.

[0094] For the deduplication sub-process, for each question dimension in the multiple question dimensions, when it is determined that the semantic information of the question dimension is consistent with the semantic information of the query result associated with the query information, the question dimension can be deleted from the multiple question dimensions to obtain the deduplicated question dimension.

[0095] For the evaluation sub-process, the target question dimensions can be determined from the question dimensions after deduplication according to the evaluation values ​​of the question dimensions after deduplication.

[0096] It should be noted that if some content has been fed back in the query results, there is no need to repeatedly ask the user related questions. Therefore, repeated questions can be avoided through the deduplication process, thereby improving the accuracy of the question dimensions. By evaluating the question dimensions, the accuracy of the target question dimensions can be made more in line with the user's query needs.

[0097] According to another embodiment of the present disclosure, the deduplication sub-process in the process of screening the target question dimensions may include: combining the second prompt information template, query information, query results and multiple question dimensions into the second prompt information, and then inputting the second prompt information into the large model to obtain the deduplicated question dimensions.

[0098] The form of the second prompt information template is similar to that of the first prompt information template. For example, the second prompt information template can be pre-configured and stored in a predetermined storage area, and read from the storage area when needed. For example, the second prompt information template includes a second thinking chain, and the second thinking chain includes multiple steps. The second thinking chain indicates that the large model performs reasoning according to the steps in the second thinking chain. For example, the multiple steps included in the second thinking chain include: generating summary information for the query result information, and determining whether the summary information is related to the question dimension. In other examples, the second prompt information template can also indicate the identity and task of the large model. The second prompt information template can also indicate the format of the input information and output information of the large model. The second prompt information template can also include examples. In this embodiment, the second prompt information template includes a second thinking chain, and the large model is guided by the steps in the second thinking chain, so that the large model can accurately perform deduplication processing, thereby avoiding repeated questioning of content that already exists in the query results.

[0099] For example, in this embodiment, the second prompt information template is as follows:

[0100] You are a robot that evaluates whether the question dimensions are consistent with the semantics of the query result answer. Your responsibility is to <query> 、 <answer>and <question dimension> to generate a logical chain of thinking about whether the current question dimension has the same semantics as the query result.

[0101] The specific thinking steps and logic are as follows:

[0102] Step 1: Deeply understand the current <query> 、 <answer>and <questioning dimension>, think deeply and reflect on <questioning dimension> to understand the current <query>The value and impact of the query are summarized and abstracted into the main contents that have been responded in the current query results.

[0103] Step 2: Starting from the given question dimension list, determine whether the content of each question dimension has been mentioned in the above step 2. If it is mentioned, it is determined that the semantics of the question dimension and the current query result are repeated, otherwise there is no repetition.

[0104] Step 3: Output the corresponding JSON data.

[0105] Below I will give you examples, and you need to learn how to generate the required chain of thought through examples.

[0106] Example: BBB.

[0107] Below I will give you an input. You need to think deeply and reflect based on what you have learned above, and finally generate the corresponding de-duplication thinking chain.

[0108] The second prompt information template is described above.

[0109] According to another embodiment of the present disclosure, the evaluation value sub-process in the process of screening the target question dimension may include: combining the third prompt information template, query information, query results and the deduplicated question dimension into the third prompt information, and then inputting the third prompt information into the big model to obtain the target question dimension.

[0110] The form of the third prompt information template is similar to the first prompt information template and the second prompt information template. For example, the third prompt information template can be pre-configured and stored in a predetermined storage area, and read from the storage area when it is needed. For example, the third prompt information template includes a third thinking chain, and the third thinking chain includes multiple steps. The third thinking chain indicates that the large model performs reasoning according to the steps in the third thinking chain. For example, the multiple steps included in the third thinking chain include: determining the evaluation value of the question dimension based on a predetermined evaluation method, and selecting the target question dimension from the question dimension after deduplication based on the evaluation value. The predetermined evaluation method may include evaluating the criticality of the question dimension affecting the current answer, and may also include evaluating the question dimension from the perspective of whether it helps to further understand the user's needs. In other examples, the third prompt information template may also indicate the identity and task of the large model. The third prompt information template may also indicate the format of the input information and output information of the large model. The third prompt information template may also include examples. In this embodiment, the third prompt information template includes a third thinking chain, and the large model is guided by the steps in the third thinking chain, so that the large model can accurately screen out more valuable question dimensions, thereby improving user satisfaction and ensuring the value and applicability of the question dimensions.

[0111] For example, in this embodiment, the third prompt information template is as follows:

[0112] You are a robot that evaluates the a priori value of question dimensions. Your responsibility is to <query> 、 <answer>and <Questioning Dimension> to generate the logical chain of thinking of "analyzing the prior value of questioning dimension".

[0113] The specific thinking steps and logic are as follows:

[0114] Step 1: Deeply understand the current <query> 、 <answer>and <questioning dimension>, think deeply and reflect on <questioning dimension> to understand the current <query>value and impact.

[0115] Step 2: Evaluate the value of the newly proposed question dimension. The specific evaluation criteria are whether the current refinement direction is a key factor affecting the answer to the question, and whether it is effective for in-depth understanding of the current problem needs.

[0116] Step 3. Finally, select the three most valuable target question dimensions. Under each target question dimension, please provide up to four options and determine whether the options are mutually exclusive.

[0117] Example: CCC.

[0118] The third prompt information template is described above.

[0119] According to another embodiment of the present disclosure, after obtaining multiple target question dimensions, the multiple target question dimensions can also be sorted, and can be sorted according to the frequency of the target question dimensions being clicked. For example, the user group to which the user who inputs the query information belongs can be determined, and then the frequency of the target question dimensions clicked by the user group during the historical search process can be determined. The target question dimensions are then sorted based on the frequency. It should be noted that sometimes the intentions of multiple users in the same user group are relatively similar. This embodiment determines the sorting of the target question dimensions through the historical operations of the user group, so that more valuable target question dimensions can be displayed preferentially.

[0120] According to another embodiment of the present disclosure, the process of determining the interaction information may include: combining the fourth prompt information template, query information and target question dimension into the fourth prompt information, and then inputting the fourth prompt information into the large model to obtain the question in the interaction information.

[0121] The form of the fourth prompt information template is similar to the first prompt information template, the second prompt information template, and the fourth prompt information template. For example, the fourth prompt information template can be pre-configured and stored in a predetermined storage area, and read from the storage area when needed. For example, the fourth prompt information template includes a fourth thinking chain, and the fourth thinking chain includes multiple steps. The fourth thinking chain indicates that the large model performs reasoning according to the steps in the fourth thinking chain. For example, the multiple steps included in the fourth thinking chain include: determining the query scenario based on the query information and the question dimension, and generating the question in the interactive information according to the query scenario. In other examples, the fourth prompt information template can also indicate the identity and task of the large model. The fourth prompt information template can also indicate the format of the input information and output information of the large model. The fourth prompt information template can also include examples. In this embodiment, the fourth prompt information template includes a fourth thinking chain, and the large model is guided by the steps in the fourth thinking chain, so that the large model can accurately output the question.

[0122] For example, in this embodiment, the fourth prompt information template is as follows:

[0123] Your task is to reconstruct the question based on the original search question, the information angle of the question, and the options of the question. Specifically, you need to follow the following steps:

[0124] Step 1: Deeply understand the current search question and the information angle of the question, analyze the key factors of the question and the user's needs, and infer the possible background or scenario when the user asks the question.

[0125] Step 2: Based on the user's question context, refer to the provided question information angles and options to form a concise and fluent question.

[0126] Example:

[0127] Input: "Search Question": "AA"; "Dimension Name": "BB"; "Option": "CC".

[0128] Output:

[0129] "Step 1": "The user asks a question DD, the key factor of this question is EE, and the user's need is FF. It is speculated that the background or scenario that the user may face is GG, so HH needs to be determined."

[0130] "Step 2": "The information angle of the question is II. Based on the current question context, we can ask questions targeting keyword JJ and guide users to supplement KK."

[0131] "Final Question": "LL".

[0132] The fourth prompt information template is described above.

[0133] It should be noted that some embodiments of the present disclosure combine a large language model and prompt engineering (PromptEngineering), and use prompt information templates to guide the output results of the large model. The large model can be a large language model (LLM), a generation model, etc., and the present disclosure does not limit the large model. Prompt information templates (such as pre-processing prompt information templates, first prompt information templates, second prompt information templates, third prompt information templates, fourth prompt information templates, and fifth prompt information templates, etc.) can all be text information described in natural language. Each prompt information template can be pre-configured and stored in a predetermined storage area, and read from the storage area when needed.

[0134] According to another embodiment of the present disclosure, after the interaction information is obtained, the interaction information can be applied in a variety of ways. In one application scenario, the interaction information can be displayed on a front-end page so that the user can make a selection based on the question dimension, thereby supplementing the query content.

[0135] For example, after the user selects some options, a new query information can be re-determined based on the options clicked by the user, and then a query can be performed based on the new query information, thereby updating the query results displayed to the user. For example, if the query information is "Chinese medicine conditioning body", the question dimension name is "gender", and the user option is "female", then the new query information can be "Chinese medicine conditioning body for women".

[0136] In addition, after the user selects an option, the option selected by the user can also be converted into natural language. For example, if the question dimension name is "gender" and the user option is "female", the converted text information can be "My gender is female", and the converted text information can be entered in the input box on the page.

[0137] According to another embodiment of the present disclosure, after the interaction information is obtained, there are multiple ways to apply the interaction information. In another application scenario, the interaction information can be used to construct training samples, and the training samples can be used to train other models, so that other trained models can be used to interact with users when resources are limited.

[0138] For example, taking the other models including two SFT models as an example, for the sake of distinction, one SFT model is referred to as the first model and the other SFT model is referred to as the second model below.

[0139] For the first model, the first model may be a generative model. During the training process, the interactive information may be precisely labeled, and then the precisely labeled data may be used as training samples to train the first model. After training, the input of the first model may include query information, historical query information, query results associated with the query information, and may also include a prompt information template applicable to the first model. The output of the first model may include question dimensions, question options, and question questions after natural language conversion of the question dimensions.

[0140] For example, the output examples of the first model are as follows: {"question dimension": "Name meaning", "question asked": "For your child's name, what meaning do you hope it will have", "options": ["smart and wise", "healthy and safe", "high moral character"], "are options mutually exclusive": "yes"}, {"question dimension": "Name origin", "question asked": "What do you hope the name will come from", "options": ["Tang poetry", "Song lyrics", "Book of Songs", "Chu Ci"], "are options mutually exclusive": "no"}.

[0141] As for the second model, the second model can be a generation model. The second model is used to generate a coherent natural language text based on the above-mentioned question dimensions, question options and question questions, so as to fill in the input box of the front-end page. For example, the output of the second model is: My baby's gender is a boy, and I hope the meaning contained is brave, strong, kind and friendly.

[0142] For example, the second model may be guided to generate natural language questions through a pre-configured prompt information template, and the prompt information template may be, for example:

[0143] Suppose you are an artificial intelligence assistant. Your main function is to analyze problems and add additional dimensions, and to provide feedback to users in a more natural language to guide users to communicate with you continuously.

[0144] The overall analysis process is as follows:

[0145] Thinking process: The first step is to combine the question and modify the guided description in natural language for each dimension based on the perspective of the questioner in the interactive guidance. The second step is to randomly select an option for each dimension, combine the question and give a natural language supplementary statement of the needs based on the perspective of the respondent in the interactive guidance.

[0146] Dimension-guided transformation: Based on the first step in the thinking process, each dimension is given a modified result after being described in natural language.

[0147] Supplementary statement of needs: Based on the second step of the thinking process, provide guidance for the supplementary statement of needs.

[0148] Here is an example:

[0149] Question: What is the difference between a quilt and a mattress?

[0150] Supplementary dimensions: [{dimension name": "Comparison aspect", "options under dimension": ["Thickness", "Material", "Cleaning"]}].

[0151] Overall analysis process:

[0152] "Thinking process": There is one dimension "comparative aspect" in the supplementary dimension. Combined with the question, a natural language description can be made, which can be changed to "Regarding the difference between quilts and mattresses, the comparative aspect you are more concerned about is". Assuming that the user selects "thickness" under the "comparative aspect" dimension, the user's own supplementary statement can be further organized, specifically "I want to know the difference between quilts and mattresses in terms of thickness".

[0153] "Dimension guides conversion": ["Regarding the difference between quilts and mattresses, the comparison aspect you are more concerned about is"].

[0154] "Supplementary statement of demand": {"Prefix":"I want to know the difference between quilt and mattress", "Dimension option combination":["The comparison aspect is thickness"], "Suffix":""}.

[0155] Note the following: The output format can be parsed using a predefined function.

[0156] Based on the above overall analysis process and examples, please analyze the following questions and supplementary dimensions:

[0157] Question: MMM.

[0158] Supplementary dimension: NNN.

[0159] The prompt information template for guiding the second model is described above.

[0160] Figure 4 It is a schematic structural block diagram of an apparatus for determining interaction information according to an embodiment of the present disclosure.

[0161] like Figure 4 As shown, the apparatus 400 for determining interaction information may include a first determining module 410 , a second determining module 420 , and a third determining module 430 .

[0162] The first determination module 410 is used to determine a plurality of question dimensions according to the query information and historical query information of the object, each question dimension including a dimension name and a plurality of options.

[0163] The second determination module 420 is used to determine a target question dimension from the multiple question dimensions according to whether the semantic information of the multiple question dimensions is consistent with the semantic information of the query result associated with the query information, and the evaluation values ​​of the multiple question dimensions.

[0164] The third determination module 430 is used to determine the interaction information according to the dimension name and multiple options in the target question dimension.

[0165] In another embodiment of the present disclosure, the above-mentioned device further includes: a historical information determination module, which is used to determine the candidate historical query information that meets the following conditions as the historical query information among the candidate historical query information: the intention of the candidate historical query information is consistent with the intention of the query information; Figure 1 The search frequency of the candidate historical query information meets the predetermined frequency condition, and the entity included in the candidate historical query information is consistent with the entity included in the query information.

[0166] In another embodiment of the present disclosure, the first determination module includes: a first combination submodule and a first input submodule. The first combination submodule is used to combine the first prompt information template, the query information, the historical query information and the query result associated with the query information into the first prompt information; the first input submodule is used to input the first prompt information into the big model to obtain multiple question dimensions; wherein the first prompt information template includes a first thought chain, and the first thought chain indicates that the big model obtains the multiple question dimensions in the following manner: determining the relationship type between the query information and the historical query information, determining whether there is a question dimension based on the relationship type, and generating question questions and options based on a predetermined refinement direction in the case of candidate question dimensions.

[0167] In another embodiment of the present disclosure, the second determination module includes: a deletion submodule and a determination submodule. The deletion submodule is used for deleting the question dimension from the multiple question dimensions for each question dimension in the plurality of question dimensions, when it is determined that the semantic information of the question dimension is consistent with the semantic information of the query result associated with the query information, to obtain the question dimension after deduplication; the determination submodule is used for determining the target question dimension from the question dimensions after deduplication according to the evaluation values ​​of each question dimension after deduplication.

[0168] In another embodiment of the present disclosure, the deletion submodule includes: a first combination unit and a first input unit. The first combination unit is used to combine the second prompt information template, query information, query results and multiple question dimensions into the second prompt information; the first input unit is used to input the second prompt information into the large model to obtain the question dimensions after deduplication; wherein the second prompt information template includes a second thought chain, and the second thought chain indicates that the large model obtains the question dimensions after deduplication in the following manner: generating summary information for the query results, and determining whether the summary information is related to the question dimensions.

[0169] In another embodiment of the present disclosure, the determination submodule includes: a second combination unit and a second input unit. The second combination unit is used to combine the third prompt information template, the query information, the query result and the question dimension after deduplication into the third prompt information; the second input unit is used to input the third prompt information into the large model to obtain the target question dimension; wherein the third prompt information template includes a third thought chain, and the third thought chain indicates that the large model obtains the target question dimension in the following manner: determining the evaluation value of the question dimension based on a predetermined evaluation method, and selecting the target question dimension from the question dimension after deduplication based on the evaluation value.

[0170] In another embodiment of the present disclosure, the third determination module includes: a second combination submodule and a second input submodule. The second combination submodule is used to combine the fourth prompt information template, the query information and the target question dimension into the fourth prompt information; the second input submodule is used to input the fourth prompt information into the large model to obtain the question in the interactive information; wherein the fourth prompt information template includes a fourth thought chain, and the fourth thought chain indicates that the large model obtains the question in the interactive information in the following manner: determining the query scenario based on the query information and the question dimension, and generating the question in the interactive information according to the query scenario.

[0171] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned method for determining interaction information.

[0172] According to an embodiment of the present disclosure, the present disclosure further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the above method for determining interaction information.

[0173] According to an embodiment of the present disclosure, the present disclosure further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method for determining the interaction information is implemented.

[0174] Figure 5 1 is a block diagram of an electronic device for implementing a method for determining interaction information of an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0175] like Figure 5 As shown, the device 500 includes a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0176] A number of components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0177] The computing unit 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 501 performs the various methods and processes described above, such as a method for determining interaction information. For example, in some embodiments, the method for determining interaction information may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the method for determining interaction information described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform the method for determining interaction information in any other appropriate manner (e.g., by means of firmware).

[0178] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0179] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0180] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0181] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0182] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0183] A computer system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship to each other.

[0184] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0185] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.< / query> < / answer> < / query> < / answer> < / query> < / query> < / answer> < / query> < / answer> < / query>

Claims

1. A method for determining interaction information, comprising: Determining multiple question dimensions according to the query information and historical query information of the object, including: combining a first prompt information template, the query information, the historical query information, and a query result associated with the query information into first prompt information; inputting the first prompt information into a large model to obtain the multiple question dimensions; each question dimension includes a dimension name and multiple options; determining a target question dimension from the multiple question dimensions according to whether the semantic information of the multiple question dimensions is consistent with the semantic information of the query result associated with the query information, and the evaluation values ​​of the multiple question dimensions; and Determining interaction information according to a dimension name and multiple options in the target question dimension; Among them, the first prompt information template includes a first chain of thought, which indicates that the large model obtains the multiple question dimensions in the following manner: determining the relationship type between the query information and the historical query information, determining whether there is a question dimension based on the relationship type, and generating question dimensions and options based on predetermined refinement directions in the case of candidate question dimensions.

2. The method according to claim 1, further comprising: The candidate historical query information that meets the following conditions among the candidate historical query information is determined as the historical query information: The intent of the candidate historical query information is consistent with the intent of the query information, the search frequency of the candidate historical query information satisfies a predetermined frequency condition, and the entity included in the candidate historical query information is consistent with the entity included in the query information.

3. The method according to any one of claims 1 to 2, wherein: The determining a target question dimension from the multiple question dimensions according to whether the semantic information of the multiple question dimensions is consistent with the semantic information of the query result associated with the query information, and the evaluation values ​​of the multiple question dimensions comprises: For each question dimension in the multiple question dimensions, when it is determined that the semantic information of the question dimension is consistent with the semantic information of the query result associated with the query information, deleting the question dimension from the multiple question dimensions to obtain a question dimension after deduplication; and The target question dimension is determined from the question dimensions after the duplicates are removed according to the evaluation values ​​of the question dimensions after the duplicates are removed.

4. The method according to claim 3, wherein: When it is determined that the semantic information of the question dimension is consistent with the semantic information of the query result associated with the query information, deleting the question dimension from the multiple question dimensions to obtain the question dimension after deduplication includes: combining the second prompt information template, the query information, the query result and the multiple question dimensions into the second prompt information; and Input the second prompt information into the large model to obtain the question dimension after deduplication; Among them, the second prompt information template includes a second chain of thought, which indicates that the large model obtains the question dimension after deduplication in the following way: generating summary information for the query result, and determining whether the summary information is related to the question dimension.

5. The method according to claim 3, wherein: Determining the target question dimension from the question dimensions after deduplication according to the evaluation values ​​of the question dimensions after deduplication includes: combining the third prompt information template, the query information, the query result and the question dimension after deduplication into the third prompt information; and Inputting the third prompt information into the large model to obtain the target question dimension; Among them, the third prompt information template includes a third chain of thinking, and the third chain of thinking indicates that the large model obtains the target question dimension in the following way: determining the evaluation value of the question dimension based on a predetermined evaluation method, and selecting the target question dimension from the deduplicated question dimension based on the evaluation value.

6. The method according to claim 1, wherein: The determining of the interaction information according to the dimension name and multiple options in the target question dimension includes: combining the fourth prompt information template, the query information and the target question dimension into fourth prompt information; and Inputting the fourth prompt information into the large model to obtain the question in the interactive information; Among them, the fourth prompt information template includes a fourth thinking chain, and the fourth thinking chain indicates that the big model obtains the questions in the interactive information in the following way: determining the query scenario based on the query information and the question dimension, and generating the questions in the interactive information according to the query scenario.

7. A device for determining interaction information, comprising: A first determination module is used to determine multiple question dimensions according to the query information and historical query information of the object, each question dimension including a dimension name and multiple options; A second determination module, configured to determine a target question dimension from the multiple question dimensions according to whether the semantic information of the multiple question dimensions is consistent with the semantic information of the query result associated with the query information, and the evaluation values ​​of the multiple question dimensions; as well as A third determination module, configured to determine interaction information according to a dimension name and multiple options in the target question dimension; Wherein, the first determining module includes: a first combining submodule, configured to combine the first prompt information template, the query information, the historical query information, and the query result associated with the query information into the first prompt information; and A first input submodule, used for inputting the first prompt information into a large model to obtain the multiple question dimensions; Among them, the first prompt information template includes a first chain of thought, which indicates that the large model obtains the multiple question dimensions in the following manner: determining the relationship type between the query information and the historical query information, determining whether there is a question dimension based on the relationship type, and generating question questions and options based on a predetermined refinement direction in the case of candidate question dimensions.

8. The apparatus according to claim 7, further comprising: The historical information determination module is used to determine the candidate historical query information that meets the following conditions among the candidate historical query information as the historical query information: The intent of the candidate historical query information is consistent with the intent of the query information, the search frequency of the candidate historical query information satisfies a predetermined frequency condition, and the entity included in the candidate historical query information is consistent with the entity included in the query information.

9. The device according to any one of claims 7 to 8, wherein: The second determining module comprises: a deleting submodule, configured to, for each question dimension in the multiple question dimensions, delete the question dimension from the multiple question dimensions when it is determined that the semantic information of the question dimension is consistent with the semantic information of the query result associated with the query information, so as to obtain a question dimension after deduplication; and The determination submodule is used to determine the target question dimension from the question dimensions after deduplication according to the respective evaluation values ​​of the question dimensions after deduplication.

10. The device according to claim 9, wherein: The deletion submodule includes: A first combining unit, configured to combine a second prompt information template, the query information, the query result, and the multiple question dimensions into second prompt information; and A first input unit, used for inputting the second prompt information into a large model to obtain a question dimension after deduplication; Among them, the second prompt information template includes a second chain of thought, which indicates that the large model obtains the question dimension after deduplication in the following way: generating summary information for the query result, and determining whether the summary information is related to the question dimension.

11. The device according to claim 9, wherein: The determination submodule comprises: A second combining unit is used to combine a third prompt information template, the query information, the query result and the question dimension after deduplication into third prompt information; and A second input unit, used to input the third prompt information into the large model to obtain the target question dimension; Among them, the third prompt information template includes a third chain of thinking, and the third chain of thinking indicates that the large model obtains the target question dimension in the following way: determining the evaluation value of the question dimension based on a predetermined evaluation method, and selecting the target question dimension from the deduplicated question dimension based on the evaluation value.

12. The device according to claim 7, wherein: The third determination module comprises: A second combining submodule, configured to combine a fourth prompt information template, the query information and the target question dimension into fourth prompt information; and A second input submodule, used for inputting the fourth prompt information into the large model to obtain the question in the interactive information; Among them, the fourth prompt information template includes a fourth thinking chain, and the fourth thinking chain indicates that the big model obtains the questions in the interactive information in the following way: determining the query scenario based on the query information and the question dimension, and generating the questions in the interactive information according to the query scenario.

13. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 6.

15. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images