Method and apparatus for generating multi-path cooperative query reply information, and electronic device

Abstract

This disclosure presents a method, apparatus, and electronic device for generating multi-path collaborative query response information. One specific implementation of the method includes: acquiring a preset execution path dataset and lexical consumption constraint information; inputting user query request information into a semantic complexity classification model to obtain query complexity information; calling hierarchical execution path data; in response to determining that the hierarchical execution path data is basic path data, executing a basic query response information generation task to obtain query response information; in response to determining that the execution path data is optimized retrieval path data, executing a retrieval query response information generation task to obtain query response information; in response to determining that the execution path data is hierarchical progressive decomposition path data, executing a hierarchical progressive decomposition query response information generation task to obtain query response information; and sending the query response information to a preset user terminal. This implementation reduces lexical consumption and the number of times device lag or query response information generation failures occur.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, specifically to a method, apparatus, and electronic device for generating multi-path collaborative query response information. Background Technology

[0002] With the groundbreaking advancements in Large Language Models (LLMs) within the field of natural language processing, various strategies have been developed to enhance the inference capabilities of LLMs in order to address complex problems. Query response generation is one such technique, where user query requests are input into the LLM to obtain query responses. Currently, the common approach for generating query responses is as follows: upon receiving a user query, the model is guided through a series of intermediate inference steps using Chain-of-Thought (CoT) prompting technology to arrive at the corresponding query response.

[0003] However, when generating multi-path collaborative query response information using the above method, the following technical problems often arise: Upon receiving a user query, the model is guided to generate a series of intermediate reasoning steps using Chain-of-Thought (CoT) prompting technology to obtain the corresponding query response information. In order to fully demonstrate its "thinking process," the model generates lengthy and detailed reasoning chains for both complex and simple questions. This directly leads to a sharp increase in the number of output tokens. Furthermore, given the limited capabilities of the device, generating a large amount of response information data can overwhelm the device, causing it to lag or fail to generate query response information.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not form prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a method, apparatus, and electronic device for generating multi-path collaborative query response information to solve one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a method for generating multi-path collaborative query response information. The method includes: in response to receiving user query request information sent by a preset user terminal, containing user terminal configuration information and user query requirement information, obtaining a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information, wherein the preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and graded progressive decomposition path data; inputting the aforementioned user query requirement information into a pre-trained semantic complexity classification model to obtain query complexity information; retrieving hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset; and in response to determining that the aforementioned hierarchical execution path data is basic path data, executing the corresponding basic path data... Based on the corresponding basic query response information generation task, query response information is generated based on the aforementioned user query requirement information and the aforementioned lexical consumption constraint information. In response to determining that the aforementioned execution path data is optimized retrieval path data, a query response information generation task corresponding to the aforementioned optimized retrieval path data is executed to perform enhanced retrieval processing on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information, thereby obtaining query response information. In response to determining that the aforementioned execution path data is tiered progressive decomposition path data, a tiered progressive decomposition query response information generation task corresponding to the aforementioned tiered progressive decomposition path data is executed to perform tiered progressive decomposition processing on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information, thereby generating query response information. The query response information is then sent to the aforementioned preset user terminal.

[0008] Secondly, some embodiments of this disclosure provide a multi-path collaborative query response information generation apparatus, the apparatus comprising: an acquisition unit configured to, in response to receiving user query request information sent by a preset user terminal, including user terminal configuration information and user query requirement information, acquire a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information, wherein the preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and graded progressive decomposition path data; an input unit configured to input the aforementioned user query requirement information into a pre-trained semantic complexity classification model to obtain query complexity information; a calling unit configured to call the hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset; and a first execution unit configured to, in response to determining that the aforementioned hierarchical execution path data is basic path data, execute the aforementioned basic path data... The system comprises the following components: a first execution unit and a second execution unit. The first execution unit is configured to generate a basic query response information based on the path data, which is determined to be optimized retrieval path data. The second execution unit is configured to generate a query response information based on the optimized retrieval path data, which is configured to perform enhanced retrieval processing on the user query information based on the lexical consumption constraint information to obtain query response information. The third execution unit is configured to generate a tiered progressive decomposition query response information based on the tiered progressive decomposition path data, which is configured to perform tiered progressive decomposition processing on the user query information based on the lexical consumption constraint information to generate query response information. The fourth execution unit is configured to send the query response information to the preset user terminal.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0011] The above-described embodiments of this disclosure have the following beneficial effects: the multi-path collaborative query response information generation method of some embodiments of this disclosure reduces the amount of token consumption and the number of times the device lags or query response information generation fails. Specifically, the reason for the large amount of token consumption and the device lag or query response information generation failure is that when the Chain-of-Thought (CoT) prompting technology guides the model to generate a series of intermediate reasoning steps to obtain the corresponding query response information, in order to fully demonstrate its "thinking process", the model generates lengthy and detailed reasoning chains for both complex and simple questions. This directly leads to a sharp increase in the number of output tokens, and under the condition of limited device capabilities, generating a large amount of response information data makes it difficult for the device to handle, causing device lag or query response information generation failure. Based on this, the multi-path collaborative query response information generation method of some embodiments of this disclosure firstly, in response to receiving user query request information sent by a preset user terminal, which includes user terminal configuration information and user query requirement information, obtains a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information. The preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and tiered progressive decomposition path data. This allows for the determination of the specific content the user needs to query and lexical consumption constraints that match the user's device capabilities, used to subsequently limit the amount of information in the generated query response information. Next, the aforementioned user query requirement information is input into a pre-trained semantic complexity classification model to obtain query complexity information. This provides query complexity information characterizing the complexity of the problem in the user query requirement information. This is used to call appropriate execution paths, avoiding the use of large amounts of data and the consumption of large amounts of lexical resources for simple problems. Then, the hierarchical execution path data corresponding to the aforementioned query complexity information is retrieved from the aforementioned hierarchical execution path dataset. Next, in response to determining that the aforementioned hierarchical execution path data is the basic path data, a basic query response information generation task corresponding to the aforementioned basic path data is executed to generate query response information based on the aforementioned user query requirement information and the aforementioned lexical consumption constraint information. Thus, when the user query requirement information is relatively simple, a query response information generation path with fewer resource calls is executed, generating shorter query response information and reducing lexical resource consumption. Then, in response to determining that the aforementioned execution path data is optimized retrieval path data, a retrieval query response information generation task corresponding to the aforementioned optimized retrieval path data is executed to perform enhanced retrieval processing on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information, obtaining query response information. Therefore, when the aforementioned user query requirement information requires retrieval of external information to provide a response, the corresponding execution path can be invoked.Then, in response to the determination that the above execution path data is hierarchical progressive decomposition path data, the hierarchical progressive decomposition query response information generation task corresponding to the above hierarchical progressive decomposition path data is executed. Based on the above lexical consumption constraint information, the above user query requirement information is processed in a hierarchical progressive manner to generate query response information. Therefore, when the complexity of the above user query requirement information is high (multi-level nested questions and the need to query external data), the hierarchical progressive decomposition path data can be called again to execute the corresponding processing method, breaking down the complex problem and ensuring the reliability of the generated query response information. Finally, the above query response information is sent to the above-mentioned preset user terminal. Also, because the complexity of the user-input query requirement information is classified, different execution paths are called for user query requirement information of different complexity levels, avoiding the use of large processing resources for both complex and simple problems, and avoiding the generation of lengthy query response information. At the same time, by combining the user's device configuration information, we can obtain word consumption constraint information that is adapted to the user's device and for different execution paths. This reduces the large amount of query response information obtained by reasoning through the use of a lot of resources in simple problems, and also reduces the device lag or query response information generation failure caused by the generated query response information exceeding the device's capacity when the device is limited. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 These are flowcharts of some embodiments of the multi-path collaborative query response information generation method according to this disclosure; Figure 2 These are schematic diagrams illustrating the structure of some embodiments of the multi-path collaborative query response information generation apparatus according to this disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1 A flowchart 100 is shown, illustrating some embodiments of a multi-path collaborative query response information generation method according to the present disclosure. This multi-path collaborative query response information generation method includes the following steps: Step 101: In response to receiving a user query request information sent by a preset user terminal containing user terminal configuration information and user query requirement information, obtain a preset execution path dataset and lexical consumption constraint information corresponding to the user terminal configuration information.

[0021] In some embodiments, the execution entity (e.g., a computing device) of the multi-path collaborative query response information generation method can, in response to receiving a user query request information sent by a preset user terminal containing user terminal configuration information and user query requirement information, obtain a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information via a wired or wireless connection. The preset user terminal can be a computing device or application used by ordinary users to submit query requests. The user configuration information can be hardware information of the electronic device (e.g., a mobile phone, computer, etc.) where the submitting computing device or application is located, representing the device's capacity to handle a single execution of data, such as total system memory and memory bandwidth utilization. The user query requirement information can be the text information of the target query entered by the user. For example, the user query requirement information could be: "I want to take my parents and children to the surrounding area for two days this weekend. Please recommend a suitable place: First, preferably within a 2-hour drive from the city center. Where would be suitable? Second, are there any attractions or activities where the elderly can easily take a walk and the children can play and have fun? Third, are there any well-regarded restaurants nearby that are suitable for family gatherings and are not expensive?" The aforementioned pre-defined hierarchical execution path dataset includes basic path data, optimized retrieval path data, and tiered progressive decomposition path data. The basic path data can serve as a routing control signal to control the execution entity's invocation of information data related to pre-defined model or program actions associated with the basic path data. Similarly, the optimized retrieval path data and the tiered progressive decomposition path data can serve as routing control signals to control the execution entity's invocation of information data related to pre-defined model or program actions associated with the optimized retrieval path data. The tiered progressive decomposition path data and the tiered progressive decomposition path data can also serve as routing control signals to control the execution entity's invocation of information data related to pre-defined model or program actions associated with the tiered progressive decomposition path data. The token consumption constraint information can represent the number of tokens consumed when processing user query requests and generating query responses. This token consumption constraint information can include basic path constraint information, optimized path constraint information, and decomposition path constraint information. The basic path constraint information can be the token consumption constraint corresponding to the execution entity's invocation of the basic path data. The optimized path constraint information can be the token consumption constraint corresponding to the execution entity's invocation of the optimized retrieval path data. The aforementioned decomposition path constraint information can be the lexical consumption constraint corresponding to the execution entity calling the tiered progressive decomposition path data. Different user-side configuration information can correspond to different lexical consumption constraint information, used to limit the generation of excessively long response information that could cause device lag when the user-side device has poor capabilities.In practice, the aforementioned execution entity can respond to a user query request sent by a preset user terminal, which includes user terminal configuration information and user query requirement information, and retrieve a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information from a preset storage device via a wired or wireless connection. The preset storage device can be a hard disk drive, solid-state drive, or similar device connected to the aforementioned execution entity.

[0022] It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other currently known or future wireless connection methods.

[0023] Step 102: Input the user's query requirements into the pre-trained semantic complexity classification model to obtain query complexity information.

[0024] In some embodiments, the execution entity can input the user query information into a pre-trained semantic complexity classification model to obtain query complexity information. This query complexity information can represent the difficulty level of the questions in the user query information, and can be "simple," "medium," or "complex." The "simple" query information may contain only a single, simple question that can be answered using pre-stored internal knowledge. For example, "Who does the Dragon Boat Festival commemorate?" The "medium" query information may contain only a single, simple question that requires external retrieval of relevant content before an answer can be given. For example, "What is the current exchange rate between the US dollar and the Chinese yuan?" The "complex" query information may contain multiple interconnected questions, some of which require external retrieval of relevant content before an answer can be given. For example, "First, check the stock price of ××, then calculate how much it would cost if I bought 100 shares, and compare the returns with the same period last year?" The pre-trained semantic complexity classification model described above can be a DistilBERT model that takes user query requirement information as input and query complexity information as output. As an example, we can first obtain at least one sample user query requirement and the corresponding sample query complexity information for each of these at least one sample user query requirement. Then, using each of these at least one sample user query requirement as input and the corresponding sample query complexity information as the expected output, we train the semantic complexity classification model.

[0025] Step 103: Retrieve the hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset.

[0026] In some embodiments, the execution entity can retrieve the hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset. In practice, when the query complexity information is "simple," the execution entity can retrieve the basic path data from the hierarchical execution path dataset as the corresponding hierarchical execution path data. When the query complexity information is "medium," the execution entity can retrieve the optimized retrieval path data from the hierarchical execution path dataset as the corresponding hierarchical execution path data. When the query complexity information is "complex," the execution entity can retrieve the tiered progressive decomposition path data from the hierarchical execution path dataset as the corresponding hierarchical execution path data.

[0027] Step 104: In response to determining the hierarchical execution path data as the basic path data, execute the basic query response information generation task corresponding to the basic path data to generate query response information based on the aforementioned user query requirement information and the aforementioned lexical consumption constraint information.

[0028] In some embodiments, the execution entity may, in response to determining the hierarchical execution path data as the basic path data, execute a basic query response information generation task corresponding to the basic path data, to generate query response information based on the user query requirement information and the lexical consumption constraint information. In practice, the execution entity may input the user query requirement information into the basic query response information generation model, limiting the number of lexical units consumed when outputting the query response information to within the lexical consumption quantity represented by the basic path constraint information in the lexical consumption constraint information, thus obtaining the query response information. The query response information may be the answer to the user's query question. The basic query response information generation model may be a general large language model (e.g., Qwen-2-7B-Instruct or Llama 3.1-8B-Instruct model).

[0029] Step 105: In response to determining that the execution path data is optimized retrieval path data, execute the retrieval query response information generation task corresponding to the optimized retrieval path data, so as to enhance the retrieval processing of the user query requirement information based on the word element consumption constraint information and obtain the query response information.

[0030] In some embodiments, the execution entity may, in response to determining that the execution path data is optimized retrieval path data, execute a retrieval query response information generation task corresponding to the optimized retrieval path data, so as to perform enhanced retrieval processing on the user query demand information based on the term consumption constraint information, and obtain query response information.

[0031] In some optional implementations of certain embodiments, the aforementioned execution entity may perform enhanced retrieval processing on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information through the following steps to obtain query response information: The first step is to generate a query-related information set based on the aforementioned user query requirements. In practice, firstly, the executing entity can input the aforementioned user query requirements into a text embedding model to obtain query vector information. Next, the text data summaries included in each text data stored in the pre-set text database are determined as each comparison summary. Then, the cosine similarity value between the aforementioned query vector information and each comparison summary is generated. Then, the obtained cosine similarity values ​​are sorted in descending order to obtain a cosine similarity value sequence. Next, the first three cosine similarity values ​​in the cosine similarity value sequence are determined as target cosine similarity values. Then, the text data corresponding to each of the three target cosine similarity values ​​is determined as query-related information. The pre-trained text embedding model can be, but is not limited to, at least one of the following: the Sentence-BERT model or the BERT model. The aforementioned query vector information can be a high-dimensional vector representation obtained by converting the aforementioned user query requirements information into vectors, output by the aforementioned text embedding model. The aforementioned pre-set text database can be an open-source database storing a large amount of publicly available information, such as the FAISS database. Each of the above text data sets may include a text data summary and the main text data. The main text data may be the main text of a publicly available article or other information. The text data summary may be a vector obtained by converting the text content summary of the main text data.

[0032] The second step involves inputting the aforementioned query-related information set into a pre-trained key information extraction model to obtain the associated key information. In practice, the executing entity can input the aforementioned query-related information set into the pre-trained key information extraction model and obtain the output information as the associated key information. The associated key information can be text information obtained by summarizing the text content of each query-related information set included in the aforementioned query-related information set. The pre-trained key information extraction model can be a neural network model that takes the query-related information set as input and the associated key information as output, such as the Anthropic Claude 3 Haiku model. As an example, at least one sample query-related information set and the sample associated key information corresponding to each sample query-related information set in the at least one sample query-related information set can be obtained first. Then, each sample query-related information set in the at least one sample query-related information set is used as input, and the sample associated key information corresponding to each sample query-related information set in the at least one sample query-related information set is used as the expected output to train the key information extraction model.

[0033] The third step involves generating query response information based on the aforementioned lexical consumption constraints, key association information, and user query requirements. In practice, the executing entity can input the key association information and user query requirements into the main language model, limiting the number of lexical units consumed when outputting the query response information to within the lexical consumption limit represented by the optimization path constraint information in the aforementioned lexical consumption constraints. The resulting output information serves as the query response information. The main language model can be a general big oracle model used to generate the final query response information, such as the GPT-4o model, the Claude 3 Opus model, or the Gemini Ultra model.

[0034] Step 106: In response to determining that the execution path data is hierarchical progressive decomposition path data, execute the hierarchical progressive decomposition query response information generation task corresponding to the hierarchical progressive decomposition path data, so as to perform hierarchical progressive decomposition processing on the user query requirement information based on the lexical consumption constraint information to generate query response information.

[0035] In some embodiments, the execution entity may, in response to determining that the execution path data is hierarchical progressive decomposition path data, execute a hierarchical progressive decomposition query response information generation task corresponding to the hierarchical progressive decomposition path data, so as to perform hierarchical progressive decomposition processing on the user query requirement information based on the lexical consumption constraint information, so as to generate query response information.

[0036] In some optional implementations of certain embodiments, the aforementioned execution entity may perform a step-by-step decomposition of the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information to generate query response information through the following steps: The first step is to perform question structure parsing on the aforementioned user query information to obtain a question parsing information set. In practice, the aforementioned execution entity can use natural language processing tools to process the aforementioned user query information to obtain the question parsing information set. The aforementioned natural language processing tools may include, but are not limited to, one of the following: spaCy, NLTK (Natural Language Toolkit). Each question parsing information in the aforementioned question parsing information set can be meaningful vocabulary information obtained after decomposing the aforementioned user query information. Each question parsing information has a corresponding category identifier. The aforementioned category identifier can identify the semantic type to which the extracted vocabulary information belongs. For example, the aforementioned category identifier can be entity type (company, person, product, etc.), numerical type (growth rate, revenue, etc.), time-limited words (latest / 2023, etc.), or logical relation words (connectors such as exceeding / and / or, etc.). The aforementioned question parsing information can be entities: electric vehicle sales (entity type), growth rate (numerical type), latest financial report (time-limited word), exceeding ×× and associated with the CEO's (Chief Executive Officer's) alma mater (logical relation word).

[0037] The second step is to generate a sequence of sub-problems based on the aforementioned problem analysis information set. In practice, the executing entity can reassemble the problem analysis information set based on a preset rule set to obtain the sequence of sub-problems. This sequence can be obtained by arranging the sub-problems according to their dependencies (each subsequent problem must depend on the answer to the previous problem). Alternatively, a sub-problem can be obtained by reassembling the problem analysis information set to obtain a single problem. The preset rule set can include custom rules and sub-problem template rules. Each custom rule can be written in Python and sets the rules for generating sub-problems. For example, a custom rule could be: each sub-problem contains only one "action + object + constraint". Each sub-problem template rule can be a template rule applied when generating a sub-problem. As an example, a sub-problem template rule could be "Under {constraints}, what is the {numerical type} of {entity}?" The resulting sub-problem could be "What was the sales revenue of electric vehicles in 2023?" The third step involves inputting each sub-problem in the aforementioned sub-problem sequence into a pre-trained information source determination model to obtain a decomposed sub-problem information sequence. Each decomposed sub-problem information sequence includes a sub-problem and an information source identifier. The information source identifier can represent whether an external information database needs to be retrieved when processing the sub-problem. The information source determination model can be a neural network model that takes sub-problems as input and decomposed sub-problem information as output, such as the BERT model, RoBERTa model, or DistilBERT model. As an example, one can first obtain at least one sample sub-problem and the sample decomposed sub-problem information corresponding to each sample sub-problem within the at least one sample sub-problem. Then, using each sample sub-problem within the at least one sample sub-problem as input and the sample decomposed sub-problem information corresponding to each sample sub-problem within the at least one sample sub-problem as the expected output, the information source determination model is trained.

[0038] The fourth step involves iteratively generating response information from the decomposed sub-problem information sequence based on the aforementioned lexical consumption constraint information, thereby obtaining the sub-response information sequence.

[0039] Fifth, based on the aforementioned lexical consumption constraint information, user query requirement information, and sub-response information sequence, query response information is generated. In practice, the executing entity can input the aforementioned sub-response information sequence and user query requirement information into the main language model, limiting the number of lexical units consumed when outputting query response information to within the lexical consumption quantity represented by the decomposition path constraint information in the aforementioned lexical consumption constraint information, and obtain the output information as query response information.

[0040] In some optional implementations of certain embodiments, the execution entity may perform iterative response information generation processing on the decomposed sub-problem information sequence based on the aforementioned lexical consumption constraint information to obtain a sub-response information sequence: The first step is to obtain an initial list of sub-response information. This initial list can include a list of responses to each sub-question; initially, it is an empty list.

[0041] The second step involves generating response information based on the initial sub-response information list and performing the following steps for the decomposed sub-problem information sequence: The first sub-step involves identifying the first sub-problem information in the sub-problem information sequence as the target sub-problem information.

[0042] The second sub-step involves generating target sub-question information based on the target sub-question information and the initial sub-response information list. This target sub-question information includes sub-query information and an information source identifier. The sub-query information can be obtained by supplementing missing information from the sub-questions in the target sub-question information. In practice, the executing entity can supplement and improve the sub-questions in the target sub-question information using text information contained in the initial sub-response information list, obtaining the supplemented sub-questions as sub-query information. Then, the sub-query information and the information source identifier in the target sub-question information are identified as the target sub-question information. For example, the initial sub-response information list could include a sub-response information such as "Company A had the highest sales volume last quarter." The target sub-question information could include the sub-question "Which is the latest mobile phone model from this company?" The supplemented sub-query information could be "Which is the latest mobile phone model from Company A?".

[0043] The third sub-step involves generating sub-response information based on the sub-query information and the lexical consumption constraint information, in response to the determination that the information source identifier in the target sub-question information meets the preset identifier conditions. The preset identifier conditions can represent the need to retrieve external data.

[0044] The fourth sub-step involves adding the sub-response information to the initial sub-response information list to update the initial sub-response information list, resulting in the updated initial sub-response information list.

[0045] The fifth sub-step involves deleting the target decomposition sub-problem information from the decomposition sub-problem information sequence to update the decomposition sub-problem information sequence, resulting in the updated decomposition sub-problem information sequence.

[0046] The sixth sub-step, in response to determining that the updated disassembly sub-problem information sequence is not empty, involves performing the response information generation step again on the updated disassembly sub-problem information sequence based on the initial sub-response information list.

[0047] Step 7: In response to the determination that the updated sub-problem information sequence is empty, the initial sub-response information list is determined as the sub-response information sequence.

[0048] In some optional implementations of certain embodiments, the aforementioned execution entity can generate sub-response information based on sub-query information and lexical consumption constraint information through the following steps: The first step is to perform vector transformation on the aforementioned subquery information to obtain the target subquestion vector. In practice, the executing entity can input the aforementioned subquery information into the aforementioned text embedding model to obtain the target subquestion vector. The target subquestion vector can be a high-dimensional vector representation output by the aforementioned text embedding model, obtained by converting the aforementioned subquery information into vectors.

[0049] The second step is to obtain the text data dataset. Each piece of text data in this dataset includes a text summary and complete text information. The text summary can be a vector derived from a summary of a text article. The complete text information can be the full text content corresponding to the text summary; for example, it could be a complete news article or a complete academic paper. The text data dataset can be various publicly available information resources stored in a public database.

[0050] The third step involves generating similarity scores between the target sub-problem vector and each text summary in the text dataset, yielding various similarity information. In practice, the executing entity can determine the cosine similarity value between the target sub-problem vector and each text summary as the similarity information. Each similarity score corresponds to a single text summary in the text dataset, i.e., to a single text data point in the text dataset.

[0051] The fourth step involves generating at least one target similarity information based on the aforementioned similarity information. Each of these target similarity information pieces corresponds to a single text data item in the aforementioned text data dataset. In practice, the executing entity can identify each similarity information piece whose similarity value is greater than a preset value as the target similarity information, thus obtaining at least one target similarity information piece.

[0052] The fifth step is to determine the complete text information in the text data corresponding to each of the above at least one target similarity information as the target association sub-information.

[0053] The sixth step is to determine the obtained target association information as a target association sub-information set.

[0054] Step 7: Input the above target-related sub-information set into the pre-trained key information extraction model to obtain target key sub-information. The above target key sub-information can be the text content obtained by summarizing the complete text information of each target-related sub-information included in the above target-related sub-information set.

[0055] Step 8: Based on the aforementioned sub-query information, the aforementioned lexical consumption constraint information, and the aforementioned target key sub-information, generate sub-response information. In practice, the aforementioned executing entity can input the aforementioned associated key information and the aforementioned user query requirement information into the main language model, limiting the number of lexical units consumed when outputting query response information to within the lexical consumption quantity represented by the decomposition path constraint information in the aforementioned lexical consumption constraint information, and obtain the output information as sub-response information. The aforementioned sub-response information can be the answer content corresponding to a sub-question. In some optional implementations of some embodiments, the aforementioned executing entity can also perform the following steps: The first step involves generating sub-response information based on the sub-query information and the lexical consumption constraint information, in response to the determination that the information source identifier in the aforementioned target sub-problem information does not meet the preset identifier conditions. In practice, the executing entity can input the sub-query information into the pre-trained basic query response information generation model, limiting the number of lexical units consumed when outputting the query response information to within the lexical consumption quantity represented by the basic path constraint information in the aforementioned lexical consumption constraint information, thus obtaining the query response information for the aforementioned target sub-problem information as the sub-query information.

[0056] Step 107: Send the query response information to the preset user terminal.

[0057] In some embodiments, the execution entity may send the query response information to the preset user terminal.

[0058] In addressing the technical problems mentioned above by adopting technical solutions, the following technical issues often arise in the application scenario: when users repeatedly input query requests in multiple rounds (e.g., users ask the same question to a large model multiple times in the same dialog box). These issues include the tendency for users to habitually omit the subject or entity nouns of their queries (i.e., the input query information is inaccurate), directly generating responses based on incomplete semantics, which can easily lead to incorrect query responses. Furthermore, semantically incomplete query information can cause misjudgments of the complexity of the user's query needs, leading to the selection of incorrect execution paths and resulting in low-quality (incorrect or partially answered) query responses, resulting in a poor user experience. Therefore, this application scenario requires the following characteristics: when users input query requests in multiple rounds, the system should incorporate historical input information to reduce the generation of incorrect query responses due to information omission. Faced with these technical problems, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the aforementioned execution entity may further perform the following steps: The first step is to obtain historical query information. This historical query information can include the current user's historical query content on the same query page.

[0059] The second step involves fusing the historical query information and the user query request information to obtain the target query information. In practice, the executing entity can input the historical query information and the user query request information into the query information rewriting model, and use the model's output as the target query information. The query information rewriting model can be a general open-source model, such as the Qwen-2.5-7B-Instruct model or the Gemma-2-9B model. The target query information can be a query question supplemented by the user query request information using the historical query information. For example, if the historical query information is "How many TV series has AA acted in?" and the user query request information is "In what year was she born?", then the rewritten target query information could be "In what year was AA born?" to avoid ignoring historical query content.

[0060] The third step involves inputting the aforementioned target query information into a pre-trained enhanced semantic complexity classification model to obtain comprehensive complexity information. This comprehensive complexity information includes query complexity information and semantic confidence values. The query complexity information can represent the difficulty level of the user's query request, such as "easy," "medium," or "complex." The semantic confidence value can represent the reliability of the generated query complexity information. The pre-trained enhanced semantic complexity classification model can be a neural network model that takes the target query information as input and the comprehensive complexity information as output (e.g., DistilBERT model, RoBERTa-base model). As an example, one can first obtain at least one sample target query and the comprehensive complexity information corresponding to each sample target query within that at least one sample target query. Then, using each sample target query as input and the comprehensive complexity information corresponding to each sample target query as the expected output, the enhanced semantic complexity classification model is trained.

[0061] Fourth, in response to the determination that the semantic confidence value included in the aforementioned comprehensive complexity information does not meet the preset semantic confidence condition, the target query information is subjected to a step-by-step progressive decomposition process based on the aforementioned lexical consumption constraint information to generate query response information. The preset semantic confidence condition can be that the semantic confidence value is greater than a pre-defined value. It should be noted that the method used to generate query response information by performing a step-by-step progressive decomposition process on the target query information based on the aforementioned lexical consumption constraint information is the same as the method used to generate query response information by performing a step-by-step progressive decomposition process on the user query requirement information based on the aforementioned lexical consumption constraint information.

[0062] Fifth, in response to determining that the semantic confidence value included in the comprehensive complexity information meets the preset semantic confidence condition, the hierarchical execution path data corresponding to the complexity information in the comprehensive complexity information is retrieved from the hierarchical execution path dataset. It should be noted that retrieving the hierarchical execution path data corresponding to the complexity information in the comprehensive complexity information from the hierarchical execution path dataset is the same method used as retrieving the hierarchical execution path data corresponding to the queried complexity information from the hierarchical execution path dataset.

[0063] Step 6: In response to determining the above-mentioned hierarchical execution path data as the base path data, perform the following steps on the above-mentioned target query information: The first sub-step involves generating intermediate query response information based on the aforementioned target query information and lexical consumption constraint information. This intermediate query response information may include intermediate response information and response confidence information. The intermediate response information may be the corresponding answer to the question in the target query information. The response confidence information may be a numerical value representing the reliability of the intermediate response information. In practice, the executing entity can input the target query information into a pre-trained basic query response information generation model, limiting the number of lexical units consumed when outputting the query response information to within the number of lexical units consumed as represented by the basic path constraint information in the lexical consumption constraint information, thus obtaining the intermediate query response information. This query response information may be the answer to the user's query question. The pre-trained basic query response information generation model may be a general-purpose large language model (e.g., Qwen-2-7B-Instruct or Llama3.1-8B-Instruct model).

[0064] The second sub-step, in response to determining that the aforementioned response confidence information does not meet the preset response confidence condition, performs a step-by-step progressive decomposition process on the aforementioned target query information based on the aforementioned lexical consumption constraint information to generate query response information. The preset response confidence condition can be that the value represented by the aforementioned response confidence information is greater than a preset value. It should be noted that the method used to perform the step-by-step progressive decomposition process on the aforementioned target query information based on the aforementioned lexical consumption constraint information to generate query response information is the same as the method used to perform the step-by-step progressive decomposition process on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information to generate query response information.

[0065] Step 7: In response to determining that the above response confidence information meets the preset response confidence condition, the intermediate response information in the above intermediate query response information is determined as the query response information.

[0066] The above-described technical solution and its related content, as an inventive point of this disclosure, solve the technical problem of "poor user experience." Factors leading to a poor user experience often include: when users repeatedly query information in multiple rounds, they tend to habitually omit the subject or entity nouns of the question (i.e., the input query information is inaccurate), directly generating responses based on incomplete semantics, which easily leads to the generation of incorrect query responses. Simultaneously, semantically missing query information can easily cause misjudgment of the complexity of the user's query needs, leading to the selection of an incorrect execution path and resulting in low-quality (incorrect or partially answered) query responses, thus causing a poor user experience. Solving these factors can improve the user experience. To achieve this, firstly, historical query information is obtained. This allows the user to obtain their historical query input, which can be used to supplement and improve the current input, reducing information omissions. Next, the historical query information and the user's query needs information are fused to obtain the target query information. Thus, the historical query information can be combined with the user's query needs information to obtain a complete and specific query question. Then, the target query information is input into a pre-trained enhanced semantic complexity classification model to obtain comprehensive complexity information, which includes query complexity information and semantic confidence value. This allows for enhanced semantic analysis of the target query information, combining not only its semantics but also multi-step instructions and logical connectors, reducing initial selection errors in execution paths based on complexity. Next, in response to the determination that the semantic confidence value included in the comprehensive complexity information does not meet the preset semantic confidence condition, the target query information is subjected to a hierarchical progressive decomposition based on the lexical consumption constraint information to generate query response information. This allows for the direct adoption of a more reliable execution path when the output semantic confidence value is below a preset range (indicating that even if judged as "simple," there is still a certain degree of complexity). Finally, in response to the determination that the semantic confidence value included in the comprehensive complexity information meets the preset semantic confidence condition, hierarchical execution path data corresponding to the complexity information is retrieved from the hierarchical execution path dataset. Then, in response to determining that the aforementioned hierarchical execution path data is the basic path data, the following steps are performed on the aforementioned target query information: First, based on the aforementioned target query information and the aforementioned lexical consumption constraint information, intermediate query response information is generated, wherein the aforementioned intermediate query response information has corresponding response confidence information. Thus, intermediate query response information with a characterizing the reliability of the intermediate query response information (i.e., response confidence information) can be obtained. Next, in response to determining that the aforementioned response confidence information does not meet the preset response confidence condition, based on the aforementioned lexical consumption constraint information, the aforementioned target query information is subjected to a hierarchical progressive decomposition process to generate query response information.Therefore, the results generated from the basic path data can be judged when the complexity level is "simple," avoiding the direct return of incorrect query response information to the user. Finally, in response to the determination that the above response confidence information meets the preset response confidence condition, the intermediate response information among the above intermediate query response information is determined as the query response information. Furthermore, by combining the user's historical input query information to supplement the current query requirement information, a more specific and accurate representation of the query question (i.e., the target query information) is obtained, reducing the error in judging the complexity of the query information due to the question. Simultaneously, semantically enhanced complexity classification of the target query information improves the accuracy of complexity classification and reduces the number of incorrect execution path selections. Moreover, when the query complexity information is represented as "simple," i.e., when executing the basic query response information generation task corresponding to the basic path data, the obtained query information is used as an intermediate result, using the confidence value as a standard to avoid directly sending low-confidence (i.e., high probability of error) query response information to the preset user terminal, thereby improving the user experience.

[0067] In addressing the technical problems mentioned above, and considering the application scenario where users query frequently trending information (e.g., movie box office figures, businesses exposed by the 315 consumer rights day), the following technical issues often arise: A large number of user-inputted queries involve currently popular trending topics, leading to significant duplication of query content across different users. Even complex questions are frequently asked repeatedly in different sessions. Repeated retrievals each time consume substantial computing resources and time to generate similar query responses, resulting in significant resource waste. Therefore, this application scenario requires the following characteristics: reducing resource waste caused by repeated queries for frequently inputted information. Faced with these technical problems, we decided to adopt the following solution: In some optional implementations of certain embodiments, the aforementioned execution entity may further perform the following steps: The first step is to obtain historical query information, historical reference response information sets, and cumulative misjudgment information. Each historical reference response set can include historical query information, historical response information, historical entity information, and historical intent information. The historical query information can be a vector derived from historical user-input queries. The historical response information can be high-quality (with high user feedback accuracy) answers generated historically by the executing entity. The historical entity information can be a vector derived from entities extracted from the historical query information. The historical intent information can be a vector derived from the information representing the user's actual purpose in asking the question extracted from the historical query information. The cumulative misjudgment information can represent the probability of low ratings from user feedback over a period of time. Each historical reference response has a corresponding quality weight. This quality weight can be a numerical value representing the number of times the historical reference response was used, ranging from 0 to 1.

[0068] The second step is to perform information fusion processing on the above historical query information and the above user query request information to obtain the target query information.

[0069] The third step involves extracting key information from the aforementioned target query information to obtain target key information. This target key information includes target query information, extracted entity information, and extracted intent information. The extracted entity information can be at least one entity extracted from the target key information. The extracted intent information can be the portion of the target key information that represents the user's actual purpose in asking the question. In practice, the executing entity can input the target query information into a pre-trained key information extraction model to obtain the target key information. The pre-trained key information extraction model can be a neural network model (e.g., BERT model, RoBERTa model) that takes the target query information as input and the target key information as output. As an example, at least one sample target query information and the sample target key information corresponding to each sample target query information can be obtained first. Then, each sample target query information in the at least one sample target query information is used as input, and the sample target key information corresponding to each sample target query information in the at least one sample target query information is used as the expected output to train the key information extraction model.

[0070] The fourth step involves vector transformation processing of the aforementioned key target information to obtain a target query vector set. This target query vector set includes query text vectors, entity vectors, and intent vectors. In practice, the executing entity can input the aforementioned key target information into the aforementioned text embedding model to obtain the target query vector set. The query text vectors can be high-dimensional vector representations output by the aforementioned text embedding model, obtained by converting the target query information into vectors. The entity vectors can be high-dimensional vector representations output by the aforementioned text embedding model, obtained by converting the extracted entity information into vectors. The intent vectors can be high-dimensional vector representations output by the aforementioned text embedding model, obtained by converting the extracted intent information into vectors.

[0071] Step 5: Based on the aforementioned historical reference response information set, perform a split-path similarity retrieval process on the aforementioned target query vector set to obtain at least one similar historical reference response. Each of the at least one historical reference response has a corresponding similarity value and quality weight. In practice, for each historical reference response in the aforementioned historical reference response information set, the executing entity can perform the following steps: First, generate a cosine similarity value between the aforementioned query text vector and the aforementioned historical query information as a query similarity value. Next, generate a cosine similarity value between the aforementioned entity vector and the aforementioned historical entity information as an entity similarity value. Then, generate a cosine similarity value between the aforementioned intent vector and the aforementioned historical intent information as an intent similarity value. Then, determine the similarity value by summing the aforementioned query similarity value, entity similarity value, and intent similarity value. Next, the executing entity can determine the similarity values ​​among the obtained similarity values ​​that are greater than a preset value as target similarity values. Then, determine the historical reference response information corresponding to each of the aforementioned at least one target similarity values ​​as similar historical reference response information.

[0072] Step 6: Based on the accumulated misjudgment information and at least one quality weight corresponding to at least one similar historical reference response, a dynamic reference threshold is generated. In practice, for each of the aforementioned quality weights, the executing entity can input the accumulated misjudgment information and the aforementioned quality weight into a preset dynamic reference threshold formula to obtain a value as an intermediate reference threshold. Then, the executing entity can use the average of the obtained intermediate reference thresholds as the dynamic reference threshold. The dynamic reference threshold can be a value generated as a reference for this application. The preset dynamic reference threshold formula can be used... Represented as: in, A pre-set threshold baseline can be used, and all dynamic thresholds will fluctuate based on this baseline. This can be the cumulative misjudgment information mentioned above. The quality weights mentioned above can be used. The above intermediate reference threshold can be used.

[0073] Step 7: Determine at least one similarity value corresponding to the at least one similar historical reference response as at least one target similarity value. Each of the at least one target similarity value corresponds to one of the at least one historical reference responses.

[0074] Step 8. In response to determining that at least one of the target similarity values ​​is greater than the dynamic reference threshold, based on the lexical consumption constraint information, at least one historical reference response information, and the user query requirement information, query response information is generated, and the target key information and the query response information are added to the historical reference response information set. In practice, firstly, the executing entity can determine the target similarity values ​​that are greater than the dynamic reference threshold as usable similarity values. Next, at least one historical reference response information corresponding to the at least one usable similarity value is determined as at least one usable reference response information. Finally, the at least one usable reference response information and the user query requirement information are input into the main language model, and the number of lexical units consumed when outputting the query response information is limited to the number of lexical units consumed as represented by the decomposition path constraint information in the lexical consumption constraint information, and the output information is used as the query response information.

[0075] The above-described technical solution and its related content, as an inventive point of this disclosure, solve the technical problem of "leading to a large waste of resources." Factors leading to this waste of resources often include: a large number of user-inputted queries involve currently popular hot topics, resulting in significant duplication of query content from different users; even complex questions are frequently asked repeatedly in different sessions. Repeated retrieval each time consumes substantial computing resources and time to generate similar query responses, leading to significant resource waste. Solving these factors can reduce resource waste. To achieve this, firstly, historical query information, historical reference response information sets, and accumulated misjudgment information are obtained. Next, the historical query information and user query requirement information are fused to obtain target query information. Thus, the user query requirement information can be supplemented by historical query information to obtain a complete and specific query question. Then, key information extraction processing is performed on the target query information to obtain target key information, which includes target query information, extracted entity information, and extracted intent information. Therefore, the entities and user query intent contained in the target key information can be obtained, which are then used to match historical reference response information with high similarity. Next, the target key information is processed by vector transformation to obtain a target query vector set, which includes query text vectors, entity vectors, and intent vectors. Then, based on the historical reference response information set, the target query vector set is processed by split-path similarity retrieval to obtain at least one similar historical reference response, where each historical reference response has a corresponding similarity value and quality weight. This yields historical reference responses with high similarity in the query question, the entities contained in the question, and the user intent. Then, based on the accumulated misjudgment information and the quality weights, a dynamic reference threshold is generated. This yields a dynamic reference threshold that combines the reliability of query response information generated using the current method within a certain time period. Finally, at least one similarity value corresponding to the at least one similar historical reference response is determined as at least one target similarity value, where each target similarity value corresponds to one historical reference response. Finally, in response to determining that at least one of the target similarity values ​​is greater than the dynamic reference threshold, query response information is generated based on the lexical consumption constraint information, the at least one historical reference response information, and the user query requirement information, and the target key information and the query response information are added to the historical reference response information set.Therefore, when at least one of the aforementioned target similarity values ​​exceeds the dynamic reference threshold, it indicates that there is relatively similar historical response information that matches the current user's query needs. The corresponding query response information is then generated by combining at least one historical reference response. This process first supplements and refines the user's query needs by incorporating historical query information, resulting in more specific and accurate target query information. Then, the information in the target query information is further analyzed and extracted to obtain the specific entities and query intent contained within it, thereby improving the accuracy of matching similar historical reference responses. Then, through multi-faceted matching of target query information, specific entities, and intent, historical reference responses with higher similarity are obtained. Simultaneously, the accuracy of the query response information generated by the current path within a certain time period and the quality weights (representing the number of uses and reliability) of the historical reference responses are combined to obtain the threshold for this execution. This threshold is used to filter at least one more reliable historical reference response from the higher similarity historical reference responses to generate the final query response information, reducing the resource waste caused by multiple searches for similar content.

[0076] The above-described embodiments of this disclosure have the following beneficial effects: the multi-path collaborative query response information generation method of some embodiments of this disclosure reduces the amount of token consumption and the number of times the device lags or query response information generation fails. Specifically, the reason for the large amount of token consumption and the device lag or query response information generation failure is that when the Chain-of-Thought (CoT) prompting technology guides the model to generate a series of intermediate reasoning steps to obtain the corresponding query response information, in order to fully demonstrate its "thinking process", the model generates lengthy and detailed reasoning chains for both complex and simple questions. This directly leads to a sharp increase in the number of output tokens, and under the condition of limited device capabilities, generating a large amount of response information data makes it difficult for the device to handle, causing device lag or query response information generation failure. Based on this, the multi-path collaborative query response information generation method of some embodiments of this disclosure firstly, in response to receiving user query request information sent by a preset user terminal, which includes user terminal configuration information and user query requirement information, obtains a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information. The preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and tiered progressive decomposition path data. This allows for the determination of the specific content the user needs to query and lexical consumption constraints that match the user's device capabilities, used to subsequently limit the amount of information in the generated query response information. Next, the aforementioned user query requirement information is input into a pre-trained semantic complexity classification model to obtain query complexity information. This provides query complexity information characterizing the complexity of the problem in the user query requirement information. This is used to call appropriate execution paths, avoiding the use of large amounts of data and the consumption of large amounts of lexical resources for simple problems. Then, the hierarchical execution path data corresponding to the aforementioned query complexity information is retrieved from the aforementioned hierarchical execution path dataset. Next, in response to determining that the aforementioned hierarchical execution path data is the basic path data, a basic query response information generation task corresponding to the aforementioned basic path data is executed to generate query response information based on the aforementioned user query requirement information and the aforementioned lexical consumption constraint information. Thus, when the user query requirement information is relatively simple, a query response information generation path with fewer resource calls is executed, generating shorter query response information and reducing lexical resource consumption. Then, in response to determining that the aforementioned execution path data is optimized retrieval path data, a retrieval query response information generation task corresponding to the aforementioned optimized retrieval path data is executed to perform enhanced retrieval processing on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information, obtaining query response information. Therefore, when the aforementioned user query requirement information requires retrieval of external information to provide a response, the corresponding execution path can be invoked.Then, in response to the determination that the above execution path data is hierarchical progressive decomposition path data, the hierarchical progressive decomposition query response information generation task corresponding to the above hierarchical progressive decomposition path data is executed. Based on the above lexical consumption constraint information, the above user query requirement information is processed in a hierarchical progressive manner to generate query response information. Therefore, when the complexity of the above user query requirement information is high (multi-level nested questions and the need to query external data), the hierarchical progressive decomposition path data can be called again to execute the corresponding processing method, breaking down the complex problem and ensuring the reliability of the generated query response information. Finally, the above query response information is sent to the above-mentioned preset user terminal. Also, because the complexity of the user-input query requirement information is classified, different execution paths are called for user query requirement information of different complexity levels, avoiding the use of large processing resources for both complex and simple problems, and avoiding the generation of lengthy query response information. At the same time, by combining the user's device configuration information, we can obtain word consumption constraint information that is adapted to the user's device and for different execution paths. This reduces the large amount of query response information obtained by reasoning through the use of a lot of resources in simple problems, and also reduces the device lag or query response information generation failure caused by the generated query response information exceeding the device's capacity when the device is limited.

[0077] Further reference Figure 2 As an implementation of the methods shown in the figures, this disclosure provides some embodiments of a plate information recognition device, which are similar to... Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0078] like Figure 2As shown, a multi-path collaborative query response information generation device 200 in some embodiments includes: an acquisition unit 201, an input unit 202, a calling unit 203, a first execution unit 204, a second execution unit 205, a third execution unit 206, and a sending unit 207. The acquisition unit 201 is configured to, in response to receiving a user query request information sent by a preset user terminal containing user terminal configuration information and user query requirement information, acquire a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information. The preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and tiered progressive decomposition path data. The input unit 202 is configured to input the aforementioned user query requirement information into a pre-trained semantic complexity classification model to obtain query complexity information. The calling unit 203 is configured to call the hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset. The first execution unit 204 is configured to, in response to determining that the aforementioned hierarchical execution path data is basic path data, execute a basic query response information generation task corresponding to the aforementioned basic path data. The first execution unit 205 is configured to generate query response information based on the user query requirement information and the lexical consumption constraint information. The second execution unit 205 is configured to execute a query response information generation task corresponding to the optimized retrieval path data in response to determining that the execution path data is optimized retrieval path data. This task enhances the retrieval of the user query requirement information based on the lexical consumption constraint information to obtain query response information. The third execution unit 206 is configured to execute a step-by-step decomposition query response information generation task corresponding to the step-by-step decomposition path data in response to determining that the execution path data is step-by-step decomposition path data. This task enhances the retrieval of the user query requirement information based on the lexical consumption constraint information to generate query response information. The sending unit 207 is configured to send the query response information to the preset user terminal.

[0079] It is understandable that the units described in the device 200 are related to the reference. Figure 1 The steps in the method described above correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the device 200 and the units contained therein, and will not be repeated here.

[0080] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0081] like Figure 3As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0082] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0083] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0084] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0085] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0086] A computer-readable medium may be contained within an electronic device or may exist independently, not assembled into the electronic device. The computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: respond to receiving a user query request information from a preset user terminal containing user terminal configuration information and user query requirement information; acquire a preset execution path dataset and lexical consumption constraint information corresponding to the aforementioned user terminal configuration information, wherein the preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and tiered progressive decomposition path data; input the aforementioned user query requirement information into a pre-trained semantic complexity classification model to obtain query complexity information; retrieve the hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset; and, in response to determining that the aforementioned hierarchical execution path data is basic path data, execute the aforementioned basic... The task of generating basic query response information corresponding to the path data generates query response information based on the aforementioned user query requirement information and the aforementioned lexical consumption constraint information. In response to determining that the aforementioned execution path data is optimized retrieval path data, a task of generating retrieval query response information corresponding to the aforementioned optimized retrieval path data is executed to perform enhanced retrieval processing on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information, thereby obtaining query response information. In response to determining that the aforementioned execution path data is tiered progressive decomposition path data, a task of generating tiered progressive decomposition query response information corresponding to the aforementioned tiered progressive decomposition path data is executed to perform tiered progressive decomposition processing on the aforementioned user query requirement information based on the aforementioned lexical consumption constraint information, thereby generating query response information. The query response information is then sent to the aforementioned preset user terminal.

[0087] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0088] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0089] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor can be described as including an acquisition unit, an input unit, a calling unit, a first execution unit, a second execution unit, a third execution unit, and a sending unit. The names of these units do not necessarily limit the unit itself; for example, an input unit can also be described as "a unit that inputs the aforementioned user query requirement information into a pre-trained semantic complexity classification model to obtain query complexity information."

[0090] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, 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), and so on.

[0091] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of technical features, but should also cover other technical solutions formed by arbitrary combinations of technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for generating multi-path collaborative query response information, comprising: In response to receiving a user query request information sent by a preset user terminal containing user terminal configuration information and user query requirement information, a preset execution path dataset and lexical consumption constraint information corresponding to the user terminal configuration information are obtained. The preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and graded progressive decomposition path data. The user query information is input into a pre-trained semantic complexity classification model to obtain query complexity information; Retrieve the hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset; In response to determining the hierarchical execution path data as the basic path data, a basic query response information generation task corresponding to the basic path data is executed to generate query response information based on the user query requirement information and the lexical consumption constraint information; In response to determining that the execution path data is optimized retrieval path data, a retrieval query response information generation task corresponding to the optimized retrieval path data is executed to perform enhanced retrieval processing on the user query requirement information based on the word consumption constraint information, and obtain query response information; In response to determining that the execution path data is stepwise progressive decomposition path data, a stepwise progressive decomposition query response information generation task corresponding to the stepwise progressive decomposition path data is executed to perform stepwise progressive decomposition processing on the user query requirement information based on the word element consumption constraint information, so as to generate query response information. The query response information is sent to the preset user terminal.

2. The method according to claim 1, wherein, The enhanced retrieval processing of the user query request information based on the lexical consumption constraint information to obtain query response information includes: Based on the user's query requirements, a query-related information set is generated; The query-related information set is input into a pre-trained key information extraction model to obtain the related key information; Based on the lexical consumption constraint information, the associated key information, and the user query requirement information, query response information is generated.

3. The method according to claim 1, wherein, The step of performing a step-by-step decomposition of the user query request information based on the lexical consumption constraint information to generate query response information includes: The user query information is parsed to obtain a question parsing information set; Based on the problem analysis information set, a sequence of sub-problems is generated; Each sub-problem in the sub-problem sequence is input into a pre-trained information source determination model to obtain a decomposed sub-problem information sequence, wherein each decomposed sub-problem information in the decomposed sub-problem information sequence includes a sub-problem and an information source identifier; Based on the lexical consumption constraint information, the decomposed sub-problem information sequence is subjected to iterative response information generation processing to obtain a sub-response information sequence; Based on the lexical consumption constraint information, the user query requirement information, and the sub-response information sequence, query response information is generated.

4. The method according to claim 3, wherein, The iterative response information generation process based on the lexical consumption constraint information, performing on the decomposed sub-problem information sequence to obtain a sub-response information sequence, includes: Get the initial list of sub-response messages; Based on the initial sub-response information list, the following response information generation steps are performed on the decomposed sub-problem information sequence: The first sub-problem information in the sub-problem information sequence is identified as the target sub-problem information; Based on the target decomposition sub-problem information and the initial sub-response information list, target sub-problem information is generated, which includes sub-query information and information source identifier; In response to the determination that the information source identifier in the target sub-question information meets the preset identifier conditions, sub-response information is generated based on the sub-query information and the lexical consumption constraint information; Add sub-response information to the initial sub-response information list to update the initial sub-response information list, and use the updated initial sub-response information list as the initial sub-response information list; The target decomposition sub-problem information is deleted from the decomposition sub-problem information sequence in order to update the decomposition sub-problem information sequence, resulting in the updated decomposition sub-problem information sequence; In response to the determination that the updated sub-problem information sequence is not empty, the response information generation step is performed again on the updated sub-problem information sequence based on the initial sub-response information list; In response to the determination that the updated sub-problem information sequence is empty, the initial sub-response information list is determined as the sub-response information sequence.

5. The method according to claim 4, wherein, The process of generating sub-response information based on sub-query information and lexical consumption constraint information includes: The sub-query information is processed by vector transformation to obtain the target sub-question vector; Obtain a text data dataset, wherein each text data item in the text data dataset includes a text summary and complete text information; The similarity generation process is performed on the target sub-problem vector and each text summary included in the text data dataset to obtain various similarity information; Based on the aforementioned similarity information, at least one target similarity information is generated, wherein each target similarity information corresponds to a text data in the text data dataset; The complete text information in the text data corresponding to each of the at least one target similarity information is determined as the target association sub-information; The obtained target association information is defined as a target association sub-information set; The target-related sub-information set is input into a pre-trained key information extraction model to obtain the target key sub-information; Based on the sub-query information, the lexical consumption constraint information, and the target key sub-information, sub-response information is generated.

6. The method according to claim 4, further comprising: In response to the determination that the information source identifier in the target sub-question information does not meet the preset identifier condition, sub-response information is generated based on the sub-query information and the lexical consumption constraint information.

7. A multi-path collaborative query response information generation device, comprising: The acquisition unit is configured to, in response to receiving a user query request information sent by a preset user terminal containing user terminal configuration information and user query requirement information, acquire a preset execution path dataset and lexical consumption constraint information corresponding to the user terminal configuration information, wherein the preset hierarchical execution path dataset includes basic path data, optimized retrieval path data, and graded progressive decomposition path data; The input unit is configured to input the user query request information into a pre-trained semantic complexity classification model to obtain query complexity information; The calling unit is configured to retrieve hierarchical execution path data corresponding to the query complexity information from the hierarchical execution path dataset. The first execution unit is configured to, in response to determining the hierarchical execution path data as the basic path data, execute a basic query response information generation task corresponding to the basic path data, so as to generate query response information based on the user query demand information and the lexical consumption constraint information; The second execution unit is configured to, in response to determining that the execution path data is optimized retrieval path data, execute a retrieval query response information generation task corresponding to the optimized retrieval path data, so as to perform enhanced retrieval processing on the user query requirement information based on the word consumption constraint information, and obtain query response information; The third execution unit is configured to, in response to determining that the execution path data is stepwise progressive decomposition path data, execute a stepwise progressive decomposition query response information generation task corresponding to the stepwise progressive decomposition path data, so as to perform stepwise progressive decomposition processing on the user query requirement information based on the word consumption constraint information, so as to generate query response information. The sending unit is configured to send the query response information to the preset user terminal.

8. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

Images