Information processing method and device and storage medium

Abstract

The invention discloses an information processing method and device and a storage medium. The method comprises the following steps: receiving question information to be replied; inputting the problem information into a target model, and performing problem planning on the problem information through the target model based on a decision retrieval algorithm to obtain a plurality of problem disassembling strategies corresponding to the problem information; the problem information is split into at least one problem set according to multiple problem splitting strategies, and dependency relationships exist among sub-problems in the problem set of the at least one problem set; and performing retrieval according to the dependency relationship among the sub-questions in the at least one question set to obtain target reply information corresponding to the question information. Through the method and the device, the problem that the accuracy of reply information is low due to the fact that the problem information is easily influenced by middle wrong steps when the problem information is retrieved by utilizing a target model in the related technology is solved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular, to an information processing method, apparatus, and storage medium. Background Art

[0002] In the current fields of natural language processing and artificial intelligence, the development and application of target models have become key driving forces for promoting technological progress. Target models have demonstrated excellent capabilities in tasks such as text generation, question answering systems, and machine translation, greatly facilitating human-computer interaction and information retrieval. However, with the continuous expansion of the model application scenarios, especially in complex problem-solving and information integration, the limitations of existing technical solutions have gradually emerged, becoming an important bottleneck restricting further improvement of model performance.

[0003] Specifically, existing technologies generally adopt iterative or chain-based schemes to handle the retrieval and answering of problem information. The iterative scheme gradually approaches the answer to the problem through multiple attempts and corrections by the model; while the chain-based scheme relies on the model to perform continuous reasoning in a series of steps to finally draw a conclusion. Although these two schemes can improve the model's problem-solving ability to a certain extent, they both face an important problem: the cumulative effect of intermediate error steps. Once an error occurs in a certain link in the reasoning or retrieval chain, this error will, like dominoes, affect each subsequent step, ultimately resulting in a significant reduction in the accuracy of the reply information generated by the model.

[0004] Regarding the problem that when using a target model to retrieve problem information in related technologies, it is easily affected by intermediate error steps, resulting in low accuracy of the reply information, no effective solution has been proposed yet. Summary of the Invention

[0005] The main purpose of this application is to provide an information processing method, apparatus, and storage medium to solve the problem that when using a target model to retrieve problem information in related technologies, it is easily affected by intermediate error steps, resulting in low accuracy of the reply information.

[0006] To achieve the above objective, according to one aspect of this application, an information processing method is provided. The method includes: receiving problem information to be replied; inputting the problem information into a target model, and performing problem planning on the problem information by the target model based on a decision retrieval algorithm to obtain multiple problem decomposition strategies corresponding to the problem information; decomposing the problem information into at least one problem set according to the multiple problem decomposition strategies, where there is a dependency relationship between sub-problems in the problem set of the at least one problem set; and performing retrieval according to the dependency relationship between sub-problems in the at least one problem set to obtain target reply information corresponding to the problem information.

[0007] Further, before splitting the problem information into at least one problem set according to multiple problem splitting strategies, the method includes: generating first candidate response information according to the problem information by a target model.

[0008] Further, retrieving the target response information corresponding to the problem information according to the dependency relationship between sub-problems in the problem set includes: retrieving at least one candidate response information according to the dependency relationship between sub-problems of the problem set in at least one problem set; determining the target response information according to the first candidate response information and the at least one candidate response information.

[0009] Further, the at least one problem set includes a first problem set. Retrieving at least one candidate response information according to the dependency relationship between sub-problems of the problem set in at least one problem set includes: determining a prompt template, where the prompt template includes prompt words and prompt examples; retrieving a second candidate response information according to the prompt template for the first problem set in the at least one problem set; repeating the step of retrieving the problem set in the at least one problem set that has not been retrieved according to the prompt template until a first preset condition is reached to obtain at least one candidate response information.

[0010] Further, the first problem set includes a first sub-problem. Retrieving a second candidate response information according to the prompt template for the first problem set in the at least one problem set includes: retrieving the response information corresponding to the first sub-problem for the first sub-problem in the first problem set according to the retrieval prompt words in the prompt template; in the case where the number of sub-problems in the first problem set is one, using the response information corresponding to the first sub-problem as the second candidate response information; in the case where the number of sub-problems in the first problem set is multiple, rewriting the second sub-problem in the first problem set according to the response information corresponding to the first sub-problem and the prompt template to obtain the rewritten second sub-problem; retrieving the response information corresponding to the rewritten second sub-problem according to the prompt template; repeating the steps of rewriting the third sub-problem in the first problem set according to the response information corresponding to the second sub-problem to obtain the response information of the rewritten third sub-problem, and rewriting the next sub-problem according to the response information of the rewritten third sub-problem to obtain the response information of the rewritten next sub-problem until a second preset condition is reached to obtain the response information corresponding to multiple rewritten sub-problems; analyzing and processing the first sub-problem, the response information corresponding to the first sub-problem, the multiple rewritten sub-problems, and the response information corresponding to the multiple rewritten sub-problems according to the prompt template to obtain the second candidate response information.

[0011] Further, rewrite the second sub-question in the first question set according to the reply information corresponding to the first sub-question and the prompt template, and the rewritten second sub-question obtained includes: analyzing the rewriting prompt words in the prompt template to determine the rewriting tasks corresponding to the rewriting prompt words; extracting the rewriting learning features of the rewritten examples in the prompt template; based on the reply information corresponding to the first sub-question, the rewriting tasks, and the rewriting learning features, rewrite the second sub-question to obtain the rewritten second sub-question.

[0012] Further, determine the target reply information according to the first candidate reply information and at least one candidate reply information, including: calculating the evaluation scores corresponding to the first candidate reply information and the evaluation scores corresponding to at least one candidate reply information according to a preset calculation mechanism; using the reply information corresponding to the highest evaluation score among the evaluation scores corresponding to the first candidate reply information and the evaluation scores corresponding to at least one candidate reply information as the target reply information.

[0013] Further, determine the target reply information according to the first candidate reply information and at least one candidate reply information, including: preprocessing the first candidate reply information to obtain the preprocessed first candidate reply information; preprocessing at least one candidate reply information to obtain the preprocessed at least one candidate reply information; extracting the first target information of the preprocessed first candidate reply information; extracting multiple target information of the preprocessed at least one candidate reply information; based on a preset fusion strategy, performing a fusion process on the first target information and the multiple target information to obtain the target reply information.

[0014] To achieve the above object, according to another aspect of the present application, an information processing device is provided. The device includes: a receiving unit, configured to receive question information to be replied; an input unit, configured to input the question information into a target model, and perform question planning on the question information by the target model based on a decision retrieval algorithm to obtain various question decomposition strategies corresponding to the question information; a splitting unit, configured to split the question information into at least one question set according to the various question decomposition strategies, wherein there are dependency relationships between the sub-questions in the at least one question set; a retrieval unit, configured to perform retrieval according to the dependency relationships between the sub-questions in the at least one question set to obtain the target reply information corresponding to the question information.

[0015] Further, the device includes: a generating unit, configured to generate a first candidate reply information according to the question information by the target model before splitting the question information into at least one question set according to the various question decomposition strategies.

[0016] Further, the retrieval unit includes: a retrieval subunit, configured to perform retrieval based on the dependency relationships between sub-questions of a question set in at least one question set, and summarize according to the generation unit to obtain at least one candidate response message; a determination subunit, configured to determine a target response message according to the first candidate response message and at least one candidate response message.

[0017] Further, the retrieval subunit includes: a first determination module, configured to determine a prompt template when the at least one question set includes a first question set, where the prompt template includes prompt words and prompt examples; a first retrieval module, configured to retrieve the first question set in the at least one question set according to the prompt template to obtain a second candidate response message; a second retrieval module, configured to repeatedly execute the step of retrieving the question sets in the at least one question set that have not been retrieved according to the prompt template until a first preset condition is met to obtain at least one candidate response message.

[0018] Further, the first retrieval module includes: a first retrieval sub-module, configured to retrieve the first sub-question in the first question set according to the retrieval prompt words in the prompt template to obtain a response message corresponding to the first sub-question; a first determination sub-module, configured to use the response message corresponding to the first sub-question as the second candidate response message when the number of sub-questions in the first question set is one; a first rewriting sub-module, configured to rewrite the second sub-question in the first question set according to the response message corresponding to the first sub-question and the prompt template to obtain a rewritten second sub-question when the number of sub-questions in the first question set is multiple; a second retrieval sub-module, configured to retrieve the rewritten second sub-question according to the prompt template to obtain a response message corresponding to the rewritten second sub-question; a second rewriting sub-module, configured to repeatedly execute the step of rewriting the third sub-question in the first question set according to the response message corresponding to the second sub-question to obtain a response message corresponding to the rewritten third sub-question, and rewriting the response message of the next sub-question according to the response message of the rewritten third sub-question until a second preset condition is met to obtain response messages corresponding to multiple rewritten sub-questions; a second determination sub-module, configured to analyze and process the first sub-question, the response message corresponding to the first sub-question, the multiple rewritten sub-questions, and the response messages corresponding to the multiple rewritten sub-questions according to the prompt template to obtain the second candidate response message.

[0019] Further, the first rewriting sub-module includes: an analysis component for analyzing the rewriting prompt words in the prompt template to determine the rewriting tasks corresponding to the rewriting prompt words; an extraction component for extracting the rewriting learning features of the rewriting examples in the prompt template; and a rewriting component for rewriting the second sub-question based on the response information corresponding to the first sub-question, the rewriting tasks, and the rewriting learning features to obtain the rewritten second sub-question.

[0020] Further, the determination subunit includes: a calculation module for calculating the evaluation scores corresponding to the first candidate response information and the evaluation scores corresponding to at least one candidate response information according to a preset calculation mechanism; and a second determination module for using the response information corresponding to the highest evaluation score among the evaluation scores corresponding to the first candidate response information and the evaluation scores corresponding to at least one candidate response information as the target response information. Further, the determination subunit includes: a first preprocessing module for preprocessing the first candidate response information to obtain the preprocessed first candidate response information; a second preprocessing module for preprocessing at least one candidate response information to obtain the preprocessed at least one candidate response information; a first extraction module for extracting the first target information of the preprocessed first candidate response information; a second extraction module for extracting multiple target information of the preprocessed at least one candidate response information; and a fusion module for performing a fusion process on the first target information and the multiple target information based on a preset fusion strategy to obtain the target response information.

[0021] According to another aspect of the present application, there is provided a computer-readable storage medium, where the computer-readable storage medium includes a stored program, and when the program runs, it controls the device where the computer-readable storage medium is located to execute any one of the information processing methods.

[0022] According to another aspect of the present application, there is provided an electronic device, including: one or more processors, a memory, and one or more programs, where one or more programs are stored in the memory and are configured to be executed by one or more processors, and the one or more programs include those for executing any one of the information processing methods.

[0023] According to another aspect of the present application, there is provided a computer program product, including computer instructions, where when the computer instructions are executed by a processor, the steps of the information processing method described in any one of the above are implemented.

[0024] In the embodiments of the present application, by receiving question information to be replied; inputting the question information into a target model, and based on a decision retrieval algorithm, the target model performs question planning on the question information to obtain multiple question decomposition strategies corresponding to the question information; splitting the question information into at least one question set according to the multiple question decomposition strategies, wherein there is a dependency relationship between sub-questions in the question set of the at least one question set; retrieving according to the dependency relationship between sub-questions in the at least one question set to obtain target reply information corresponding to the question information, which solves the technical problem that when using the target model to retrieve question information, it is easily affected by intermediate incorrect steps, resulting in a low accuracy rate of the reply information.

[0025] In the present application, the received question information input by the user is input into a target model. The target model constructs multiple question decomposition strategies corresponding to the question information based on a decision retrieval algorithm. According to the question decomposition strategies, the question information is split into at least one question set. That is, by constructing multiple question decomposition strategies, the target model does not rely on a single parsing path, but understands the question from multiple perspectives. This can reduce errors caused by the limitations of a single strategy. The target model performs logical analysis based on the relevance between sub-questions in the question set and finally retrieves the reply information. The target model not only decomposes the question but also performs logical analysis on the relevance between sub-questions, which means that the target model can understand how different parts of the question are related to each other, thereby reducing errors caused by isolated analysis. Due to having multiple question decomposition strategies, even if an error occurs in a certain step, other question decomposition strategies can still correctly guide the target model to the correct answer. Thus, the technical effect of improving the accuracy of information processing of the target model is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0027] Figure 1 A hardware structure block diagram of a computer terminal for implementing an information processing method is shown;

[0028] Figure 2 is a flowchart of an information processing method provided according to an embodiment of this application;

[0029] Figure 3 is a flowchart of retrieving candidate reply information provided according to an embodiment of this application;

[0030] Figure 4 is a schematic diagram of an information processing method provided according to an embodiment of this application;

[0031] Figure 5 It is a schematic diagram of an information processing device provided according to an embodiment of the present application;

[0032] Figure 6 It is a structural block diagram of an electronic device according to an embodiment of the present application. Specific Embodiments

[0033] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data used in appropriate cases can be interchanged so that the embodiments of the present application described here can be implemented in an order different from those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0035] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0036] Embodiment 1

[0037] According to an embodiment of the present application, an embodiment of a method for information processing is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from that here.

[0038] The method embodiment provided in Embodiment 1 of the present application can be executed on a mobile terminal, a computer terminal or a similar computing device.Figure 1 The following shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing an information processing method. As Figure 1 shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ……, 102n in the figure) (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only illustrative and does not limit the structure of the above-mentioned electronic device. For example, the computer terminal 10 may further include more or fewer components than those Figure 1 shown, or have a different configuration from that Figure 1 shown.

[0039] It should be noted that the above one or more processors 102 and / or other data processing circuits are generally referred to as "data processing circuits" in this article. The data processing circuit may be embodied in software, hardware, firmware, or any combination thereof, in whole or in part. In addition, the data processing circuit may be a single independent processing module, or be incorporated in whole or in part into any one of the other elements in the computer terminal 10 (or mobile device). As involved in the embodiments of the present application, the data processing circuit is a processor control (such as the selection of a variable resistance terminal path connected to an interface).

[0040] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the information processing method in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implements the above-mentioned information processing method. The memory 104 may include a high-speed random access memory, and may further include a non-volatile memory, such as one or more magnetic storage devices, a flash memory, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely provided with respect to the processor 102, and these remote memories can be connected to the computer terminal 10 through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0041] The transmission device 106 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one example, the transmission device 106 may be a Radio Frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0042] The display may be, for example, a touch-screen Liquid Crystal Display (LCD), which enables the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0043] Under the above operating environment, the present application provides an Figure 2 information processing method as shown in Figure 2 It is a flowchart of the information processing method provided by the embodiments of the present application.

[0044] Step S201, receive the question information to be replied.

[0045] Optionally, the above-mentioned question information may be a request information submitted by the user to the system, which requires the system to understand and process it and then give a reply or answer. It may contain the knowledge points that the user hopes to obtain, the problem description or specific query. The question information is the starting point of the entire information processing process and determines the direction and content of subsequent processing.

[0046] For example, if the user enters in the search box or dialogue interface "Among the rivers originating from the Qinghai-Tibet Plateau, how many tributaries with a drainage area of more than 10,000 square kilometers are there in the river with the largest drainage area?", then this sentence is the "question information". The question information can be an open question, a query for specific information, or a complex instruction that requires analysis and understanding.

[0047] Step S202, input the question information into the target model, and based on the decision retrieval algorithm, the target model performs question planning on the question information to obtain multiple question decomposition strategies corresponding to the question information.

[0048] Optionally, the above decision retrieval algorithm is an algorithm used to guide the target model to efficiently and strategically process and break down complex problems, and to determine which information to retrieve to assist in generating answers. The decision retrieval algorithm combines decision theory and information retrieval techniques, enabling the target model, when faced with complex problems, not only to rely on direct knowledge retrieval but also to determine the most effective information acquisition path through internal logical reasoning and strategic planning. The core of the decision retrieval algorithm lies in regarding the problem information solution as a decision-making process, in which the target model needs to evaluate the advantages and disadvantages of different problem breakdown strategies and potential information retrieval directions.

[0049] For example, the decision retrieval algorithm can be Monte Carlo tree search, which can be used for strategy optimization in games and planning problems. The basic idea of Monte Carlo tree search is to use random simulation in the search space to estimate the potential value of each action in order to gradually approach the appropriate decision.

[0050] Optionally, based on the decision retrieval algorithm for problem information, problem planning can be performed to obtain multiple problem breakdown strategies corresponding to the problem information. The above problem breakdown strategies are used to decompose complex or ambiguous queries into a series of more specific and tractable sub-problems after receiving the problem information input by the user. By performing problem planning on the problem information to obtain multiple problem breakdown strategies, the problem information can be understood and answered more effectively. By decomposing the problem information into a series of sub-problems, the target model can process these sub-problems one by one, thereby obtaining a more accurate and comprehensive answer. The above problem breakdown strategies not only include how to split the problem but also involve the logical relationship and processing order between the split sub-problems, ensuring the coherence and logic of the entire problem-solving process.

[0051] For example, the problem information proposed by the user to the system is "Among the rivers originating from the Qinghai-Tibet Plateau, how many tributaries with a drainage area of more than 10,000 square kilometers does the river with the largest drainage area have?", which is a complex problem involving multiple aspects of knowledge and requires the system to deeply understand and provide detailed answers from different perspectives. After receiving this problem, the target model designs the following three problem breakdown strategies, which are respectively:

[0052] The first problem breakdown strategy is that each sub-problem is progressive. Start by retrieving all the main rivers originating from the Qinghai-Tibet Plateau, then retrieve which river has the largest drainage area among these rivers, then retrieve what tributaries this river has, and finally retrieve the number of tributaries with a drainage area of more than 10,000 square kilometers.

[0053] The second problem decomposition strategy is as follows: The first sub-problem is required to first explore the relationship between the distribution law of tributaries and the basin area, and understand the geographical basis of the river system structure. The second sub-problem further determines which tributaries of the river with the largest basin area are. The third sub-problem confirms how many tributaries with a basin area of more than 10,000 square kilometers are among these tributaries.

[0054] The third problem decomposition strategy is as follows: The first sub-problem is required to first discuss the definition, measurement method, and data source of the basin area, providing a scientific basis for the next comparison. Then, the second sub-problem can list the main tributaries of the river with the largest river area, and understand the composition of its river system network. Finally, the third sub-problem obtains the number of tributaries with a basin area of more than 10,000 square kilometers among the tributaries.

[0055] Step S203, according to multiple problem decomposition strategies, split the problem information into at least one problem set, where there is a dependency relationship between the sub-problems in the problem set of the at least one problem set.

[0056] Split the complex problem (problem information) into multiple sub-problem sets. This process requires identifying the dependencies between sub-problems, that is, some sub-problems may require the results of other sub-problems. In this way, subsequent sub-problems can be solved one by one. This method helps to process complex problems more systematically and efficiently, reduce complexity, and improve the accuracy of answering problem information.

[0057] For example, the problem information proposed by the user to the system is "Among the rivers originating from the Qinghai-Tibet Plateau, how many tributaries with a basin area of more than 10,000 square kilometers does the river with the largest basin area have?" After receiving this problem according to the above target model, three problem decomposition strategies are designed to split the problem information into at least one problem set.

[0058] According to the first problem decomposition strategy, split the problem information to obtain problem set 1. Problem set 1 includes: The first sub-problem is "Which main rivers originate from the Qinghai-Tibet Plateau?", the second sub-problem is "Among these rivers, which one is the river with the largest basin area?", the third sub-problem is "What are the tributaries of the river with the largest basin area?", and the fourth sub-problem is "How many tributaries with an area of more than 10,000 square kilometers are there among these tributaries?"

[0059] According to the first problem decomposition strategy, split the problem information to obtain problem set 2. Problem set 2 includes: The first sub-problem is "For the rivers originating from the Qinghai-Tibet Plateau, how are their tributaries distributed and what is the relationship with the basin area?", the second sub-problem is "Specifically for the river with the largest basin area, how to count the total number of its tributaries?", and the third sub-problem is "How many tributaries with a basin area of more than 10,000 square kilometers are there among the tributaries?"

[0060] The problem information is disassembled according to the first problem disassembling strategy, and the obtained problem set 3. The problem set 3 includes: the first sub-problem is "How to measure and compare the basin areas of rivers?", the second sub-problem is "What are the main tributaries of the river that originates from the Qinghai-Tibet Plateau and has the largest basin area?", and the third sub-problem is "How many tributaries with a basin area of more than 10,000 square kilometers are there among the tributaries?".

[0061] Step S204, retrieve according to the dependency relationship between the sub-problems in at least one problem set, and obtain the target reply information corresponding to the problem information.

[0062] Optionally, the system will retrieve and answer sequentially or in parallel according to the dependency relationship of the sub-problems in the problem set. The answer to the current sub-problem can be used as the context input for the next sub-problem, ensuring the coherence and depth of the information.

[0063] For example, based on the first problem disassembling strategy in step S203 for retrieval, the execution process of retrieval according to the dependency relationship between the sub-problems in the problem set is as follows:

[0064] Retrieve and answer the first sub-problem "What are the main rivers that originate from the Qinghai-Tibet Plateau?". This step lists all the main rivers that originate from the Qinghai-Tibet Plateau, including the Yangtze River, the Yellow River, the Mekong River (Lancang River), the Yarlung Zangbo River, the Indus River, etc.; based on the answer to the first sub-problem, retrieve and answer the second sub-problem "Among these rivers, which one is the river with the largest basin area?". This step can further determine the river of interest, that is, which river has the largest basin area; combining the answers to the first two sub-problems, retrieve and answer the third sub-problem "What are the tributaries of the river with the largest basin area?". This step can retrieve what the tributaries of this river are; combining the first three sub-problems and the answers to the first three sub-problems, retrieve and answer the fourth sub-problem "How many tributaries with an area of more than 10,000 square kilometers are there among these tributaries?". This step can obtain the final answer of 49.

[0065] The information processing method provided by the embodiments of this application receives problem information to be replied; inputs the problem information into a target model, and the target model performs problem planning on the problem information based on a decision retrieval algorithm to obtain multiple problem decomposition strategies corresponding to the problem information; decomposes the problem information into at least one problem set according to the multiple problem decomposition strategies, where there are dependency relationships between sub-problems in the problem set of the at least one problem set; performs retrieval according to the dependency relationships between sub-problems in the at least one problem set to obtain target reply information corresponding to the problem information, solving the technical problem that when using the target model to retrieve problem information, it is easily affected by intermediate error steps, resulting in a low accuracy rate of the reply information. In this application, the received problem information input by the user is input into the target model. The target model constructs multiple problem decomposition strategies corresponding to the problem information based on the decision retrieval algorithm. The target model further reasons about the problem decomposition strategies, decomposes the problem information into at least one problem set, and the target model performs logical analysis according to the relevance between sub-problems in the problem set, and finally retrieves the reply information, thus achieving the technical effect of improving the accuracy of information processing of the target model.

[0066] Optionally, in order to generate more accurate target reply information, in the information processing method provided by the embodiments of this application, before decomposing the problem information into at least one problem set according to the multiple problem decomposition strategies, the method includes: generating a first candidate reply information by the target model according to the problem information.

[0067] Optionally, the above-mentioned first candidate reply information refers to attempting to directly generate a preliminary reply information that may contain partial answers by using the intrinsic knowledge of the target model before problem decomposition and in-depth retrieval. For example, after receiving the problem information "Among the rivers originating from the Qinghai-Tibet Plateau, how many tributaries of the river with the largest drainage area have a drainage area exceeding 10,000 square kilometers?", the target model will immediately use its intrinsic reasoning and language generation capabilities to generate the first candidate reply information "Among the tributaries of the river with the largest drainage area among the rivers originating from the Qinghai-Tibet Plateau, the number of tributaries with a drainage area of more than 10,000 square kilometers is 49." Although the above-mentioned first candidate reply information provides the correct answer, it does not delve into the details of the problem or provide comprehensive information.

[0068] In summary, by generating the first candidate reply information through the above steps, it can quickly respond to the user's problem, and at the same time can also be used as a reference reply answer for subsequent problem decomposition and retrieval to help obtain more accurate reply information.

[0069] Optionally, in order to obtain more comprehensive reply information, in the information processing method provided by the embodiments of this application, performing retrieval according to the dependency relationships between sub-problems in the problem set to obtain target reply information corresponding to the problem information includes:

[0070] In the first step, retrieve based on the dependency relationships among the sub-questions of the question sets in at least one question set to obtain at least one candidate response message.

[0071] For example, retrieve the three question sets in step S203 to obtain three candidate response messages. Candidate response message 1 is "Among the rivers originating from the Qinghai-Tibet Plateau, the Yangtze River is the river with the largest drainage area, reaching approximately 1.8 million square kilometers, exceeding other rivers. Among the tributaries of the Yangtze River, the number of tributaries with a drainage area of over 10,000 square kilometers is 49 tributaries." Candidate response message 2 is "The Yangtze River has numerous tributaries. The main tributaries include the Yalong River, Min River, Jialing River, Wu River, Xiang River, Gan River, etc. These tributaries form a vast water system network of the Yangtze River, not only increasing its drainage area but also having an important impact on the water volume and water quality of the Yangtze River. Among them, the number of tributaries with a drainage area exceeding 10,000 square kilometers is 49." Candidate response message 3 is "According to the analysis of geographic information systems and the statistics of hydrological data, the total number of tributaries of the Yangtze River exceeds 700, of which there are approximately 100 first-order tributaries. These tributaries are widely distributed in the Yangtze River basin, forming a vast and complex water system network from the source in the Qinghai-Tibet Plateau to the estuary in the East China Sea." It can be found that candidate response message 1 and candidate response message 2 provide correct response messages, but candidate response message 3 does not give a correct response message.

[0072] In the second step, determine the target response message based on the first candidate response message and at least one candidate response message.

[0073] For example, based on the first candidate response message "Among the tributaries of the river with the largest drainage area among the rivers originating from the Qinghai-Tibet Plateau, the number of tributaries with a drainage area of over 10,000 square kilometers is 49.", and the two correct candidate response messages mentioned above, all candidate response messages can be integrated to ensure that the final response is comprehensive, accurate, and clearly structured. The final target response message is "Among the rivers originating from the Qinghai-Tibet Plateau, the Yangtze River is the river with the largest drainage area, with a drainage area reaching approximately 1.8 million square kilometers, far exceeding other rivers. The water system of the Yangtze River is complex and has numerous tributaries. The main tributaries include the Yalong River, Min River, Jialing River, Wu River, Xiang River, Gan River, etc. These tributaries not only enrich the water volume of the Yangtze River but also cover a vast area from the source in the Qinghai-Tibet Plateau to the estuary in terms of geographical distribution, forming a vast and delicate water network system. According to the analysis of geographic information systems and the statistics of hydrological data, the total number of tributaries of the Yangtze River exceeds 700, of which there are approximately 100 first-order tributaries; particularly noteworthy is that the number of tributaries with a drainage area exceeding 10,000 square kilometers is 49. These large tributaries contribute significantly to the water volume of the Yangtze River and are also key areas for the protection of ecological diversity within the basin."

[0074] In summary, the target response information is obtained through the above steps, ensuring a comprehensive answer to the question information, thereby obtaining a more accurate and comprehensive response information.

[0075] To retrieve more efficiently according to the dependency relationships among sub-questions in a question set, optionally, as Figure 3 shown, in the information processing method provided in an embodiment of the present application, at least one question set includes a first question set. Retrieving according to the dependency relationships among sub-questions in the question sets in at least one question set, the obtained at least one candidate response information includes:

[0076] Step S301: Determine a prompt template, where the prompt template includes a prompt word and a prompt example.

[0077] Optionally, the above prompt template may be a prompt template screened from a template library according to question information. The prompt template is a structured instruction for guiding a target model to generate a target response answer. The template may include a prompt word and a prompt example. The prompt template designs specific prompt words and guiding examples according to different processing stages (such as direct retrieval, question rewriting, in-depth analysis, etc.) to guide the target model to generate a more accurate and demand-compliant answer.

[0078] For example, for the question information "Among the rivers originating from the Qinghai-Tibet Plateau, how many tributaries with a drainage area of more than 10,000 square kilometers does the river with the largest drainage area have?", the prompt template includes: the prompt word "You are an expert in answering questions. I will give you some background, which may or may not be relevant to the question. Please answer the question 'Among the rivers originating from the Qinghai-Tibet Plateau, how many tributaries with a drainage area of more than 10,000 square kilometers does the river with the largest drainage area have?' according to the context." The prompt template includes at least one specific retrieval case, showing that for a river with a very large drainage area, using a geographic information system and hydrological data to count the total number of its tributaries, and by analyzing the river network structure and the confluence points of the tributaries, a relatively accurate number of tributaries can be obtained.

[0079] Step S302: Retrieve the first question set in at least one question set according to the prompt template to obtain a second candidate response information.

[0080] For example, based on the prompt template determined in step S301, the target model will retrieve the first question set and generate a second candidate response message. Based on the prompt words and guiding examples, detailed answers closely related to the question information can be provided, including which rivers originate from the Qinghai-Tibet Plateau, which river has the largest drainage area, and what tributaries does this river have, etc. The obtained second candidate response message is "Among the rivers originating from the Qinghai-Tibet Plateau, the Yangtze River has the largest drainage area, reaching approximately 1.8 million square kilometers, exceeding other rivers. There are 49 tributaries of the Yangtze River with a drainage area of more than 10,000 square kilometers."

[0081] Step S303: Repeat the step of retrieving the unretrieved question sets in at least one question set according to the prompt template until a first preset condition is met, and obtain at least one candidate response message.

[0082] Optionally, the above first preset condition may be that all question sets have been retrieved, or the number of retrievals reaches a preset threshold (for example, 2 times), or the depth of the search tree in a single question set (for example, 5 times), where the depth of the search tree refers to the number of steps or levels required in the process of solving a problem or making a decision. For example, when the first preset condition is that the number of retrieved question sets reaches 2, the retrieval stops, and two candidate response messages can be obtained. The setting of the first preset condition can effectively control resource consumption, avoid endless retrieval or generation processes, and ensure the efficiency and feasibility of information processing.

[0083] In summary, the above steps combined with the prompt template for retrieval enable the entire information processing process to more efficiently and accurately meet the user's information needs, not only accelerating the information retrieval speed but also ensuring the depth and quality of the information.

[0084] To obtain a target response message with higher quality, optionally, in the information processing method provided in the embodiments of the present application, the first question set includes a first sub-question. Retrieving the first question set in at least one question set according to the prompt template to obtain a second candidate response message includes:

[0085] First step: Retrieve the first sub-question in the first question set according to the retrieval prompt words in the prompt template to obtain the response message corresponding to the first sub-question.

[0086] For example, for the question information "Among the rivers originating from the Qinghai-Tibet Plateau, how many tributaries of the river with the largest drainage area have a drainage area exceeding 10,000 square kilometers?", the first sub-question is "Which major rivers originate from the Qinghai-Tibet Plateau?", the second sub-question is "Among these rivers, which one has the largest drainage area?", the third sub-question is "What are the tributaries of the river with the largest drainage area?", and the fourth sub-question is "How many tributaries with an area exceeding 10,000 square kilometers are there among these tributaries?". For the retrieval prompt for the first sub-question "Which major rivers originate from the Qinghai-Tibet Plateau?" is "You are an expert in answering questions. I will give you some background information, which may or may not be relevant to the question. Please answer the question 'Which major rivers originate from the Qinghai-Tibet Plateau?' based on the context." According to this retrieval term, the reply information corresponding to the first sub-question retrieved is "The Qinghai-Tibet Plateau is the source of many major rivers in Asia, including the Yangtze River, the Yellow River, the Mekong River, the Yarlung Zangbo River, the Indus River, etc. These rivers are geographically known as the 'Water Tower of Asia' because they nourish the vast land downstream."

[0087] In the second step, when the number of sub-questions in the first question set is one, the reply information corresponding to the first sub-question is used as the second candidate reply information.

[0088] Optionally, if there are no other sub-questions subsequently, then the reply information corresponding to the first sub-question obtained in the first step will be directly used as the second candidate reply information. Then, based on the first candidate reply information and the second candidate reply information, the target reply information can be determined.

[0089] In the third step, when the number of sub-questions in the first question set is multiple, the second sub-question in the first question set is rewritten according to the reply information corresponding to the first sub-question and the prompt template, resulting in the rewritten second sub-question.

[0090] For example, based on the reply information corresponding to the first sub-question in the first step, the second sub-question "Among these rivers, which one has the largest drainage area?" is rewritten to obtain the rewritten second sub-question "Among the Yangtze River, the Yellow River, the Mekong River, the Yarlung Zangbo River, and the Indus River, which one has the largest drainage area?".

[0091] In the fourth step, the rewritten second sub-question is retrieved according to the prompt template to obtain the reply information corresponding to the rewritten second sub-question.

[0092] For example, the rewritten second sub-question is "Among the Yangtze River, the Yellow River, the Mekong River, the Yarlung Zangbo River, and the Indus River, which one is the river with the largest drainage area?" The corresponding prompt word for the prompt template can be "You are an expert in answering questions. I will give you some background information, which may or may not be relevant to the question. Please answer the question 'Among the Yangtze River, the Yellow River, the Mekong River, the Yarlung Zangbo River, and the Indus River, which one is the river with the largest drainage area?' based on the context." After retrieving the rewritten second sub-question based on the prompt word, the corresponding reply information for the second sub-question is "Among the above-mentioned rivers, the drainage area of the Yangtze River reaches approximately 1.8 million square kilometers, making it the river with the largest drainage area and one of the important rivers in China and even Asia."

[0093] Step 5: Repeat the steps of rewriting the third sub-question in the first question set according to the reply information corresponding to the second sub-question to obtain the reply information for the rewritten third sub-question, and rewriting the next sub-question according to the reply information for the rewritten third sub-question to obtain the reply information for the rewritten next sub-question until the second preset condition is reached, and obtain the reply information corresponding to multiple rewritten sub-questions.

[0094] Optionally, the above-mentioned second preset condition can be reaching a pre-set retrieval depth. For example, if the pre-set decision tree depth is 3, then after 3 rounds of retrieval and generation, regardless of whether a complete answer is obtained, the retrieval will stop. It can also be when the consumed computing resources or time reach the pre-set maximum value. Considering the computing cost and efficiency, an upper limit on resource usage is often set during the retrieval process. Once this upper limit is reached or exceeded, even if the answer is not fully determined, the retrieval must stop and an answer is generated based on the existing information. It can also be when the pre-set key information or a specific type of answer is obtained. For example, if the original question asks for the number of tributaries with a drainage area exceeding 10,000 square kilometers in the rivers originating from the Qinghai-Tibet Plateau and having the largest drainage area, once this information is retrieved, the retrieval can be terminated and the answers can be summarized.

[0095] For example, according to the reply information corresponding to the second sub-question in the fourth step, rewrite the third sub-question in the first question set, "What are the tributaries of the river with the largest drainage area?", and the rewritten third sub-question is "What are the tributaries of the Yangtze River?". The reply information for the rewritten third sub-question is "The Yangtze River has many tributaries. The main tributaries include the Yalong River, Min River, Jialing River, Han River, Wu River, Xiang River, etc. These tributaries cover most of the Yangtze River Basin." According to the reply information corresponding to the rewritten third sub-question, rewrite the fourth sub-question, and the reply information for the rewritten fourth sub-question is "Among the tributaries of the Yangtze River, there are 49 tributaries with a drainage area exceeding 10,000 square kilometers, including the Yalong River, Min River, Jialing River, etc." After the above steps, the reply information corresponding to the first sub-question, the rewritten sub-questions, and the reply information corresponding to the rewritten sub-questions are obtained.

[0096] Step 6: Analyze and process the first sub-question, the reply information corresponding to the first sub-question, multiple rewritten sub-questions, and the reply information corresponding to multiple rewritten sub-questions according to the prompt template to obtain the second candidate reply information.

[0097] Optionally, after all sub-questions have been replied, the first sub-question, the reply information corresponding to the first sub-question, multiple rewritten sub-questions, and the reply information corresponding to multiple rewritten sub-questions can be comprehensively analyzed and processed according to the prompt template and integrated into a structured and comprehensive second candidate reply information. The technical effect of this step is that through information integration, the final reply information not only comprehensively covers all aspects of the user's question, but also has a clearer structure, which is convenient for users to understand and use subsequently.

[0098] For example, by analyzing and processing the first sub-question and the reply information corresponding to the first sub-question in the first step above, the rewritten second sub-question in the third step, the rewritten reply information in the fourth step, and the rewritten third sub-question, the reply information corresponding to the rewritten third sub-question, the rewritten fourth sub-question, and the reply information corresponding to the rewritten fourth sub-question in the fifth step, the second candidate reply information can be obtained as "Among the rivers originating from the Qinghai-Tibet Plateau, the Yangtze River has the largest drainage area, reaching approximately 1.8 million square kilometers, exceeding other rivers. There are 49 tributaries of the Yangtze River with a drainage area of more than 10,000 square kilometers."

[0099] In summary, by combining the above steps with the use of the prompt template, through gradually refining the questions, deeply mining the information, and integrating the reply information, the entire information processing process not only quickly responds to the user's needs, but also generates high-quality reply information, optimizing the user experience.

[0100] Optionally, in order to improve the accuracy of retrieval, in the information processing method provided in the embodiments of the present application, the second sub-question in the first question set is rewritten according to the reply information corresponding to the first sub-question and the prompt template, and the rewritten second sub-question obtained includes:

[0101] First step, analyze the rewriting prompt words in the prompt template to determine the rewriting tasks corresponding to the rewriting prompt words.

[0102] Optionally, the above-mentioned rewriting prompt words refer to specific words or phrases used to guide and indicate changes in the question expression during the rewriting of the sub-question. These prompt words help the target model adjust the focus of the question, so as to obtain more accurate, in-depth or extensive reply information. For example, if the rewriting prompt word includes "specific application cases", the rewriting task is to require providing specific examples related to the reply information corresponding to the first sub-question.

[0103] Second step, extract the rewriting learning features of the rewritten examples in the prompt template.

[0104] Optionally, the rewritten examples usually include a set of optimized question expressions. By analyzing these rewritten examples, key features that can guide question rewriting can be extracted, such as the use of terms in a specific field, improvement of the question structure, etc.

[0105] Third step, rewrite the second sub-question based on the reply information corresponding to the first sub-question, the rewriting task, and the rewriting learning features to obtain the rewritten second sub-question.

[0106] For example, the original question set is the first sub-question "Which major rivers originate from the Qinghai-Tibet Plateau?", the second sub-question is "Among these rivers, which one has the largest drainage area?", the third sub-question is "What are the tributaries of the river with the largest drainage area?", and the fourth sub-question is "How many tributaries with an area of more than 10,000 square kilometers are there among these tributaries?". The reply information corresponding to the first sub-question is "The Qinghai-Tibet Plateau is the source of many major rivers in Asia, including the Yangtze River, the Yellow River, the Mekong River, the Yarlung Zangbo River, the Indus River, etc. These rivers are known as the 'Water Tower of Asia' in geography because they nourish the vast land downstream." The rewriting task is "Combined with the known information that the Qinghai-Tibet Plateau is the source of rivers such as the Yangtze River, the Yellow River, the Mekong River, the Yarlung Zangbo River, and the Indus River, rewrite the second sub-question." The rewriting learning features are the learning features extracted from the rewritten examples, and the rewritten examples emphasize the transformation from principle to specific application. Then, the finally obtained rewritten second sub-question is "Among the Yangtze River, the Yellow River, the Mekong River, the Yarlung Zangbo River, and the Indus River, which one has the largest drainage area?"

[0107] In summary, by rewriting the second sub-question based on the rewriting prompt words, rewriting examples in the prompt template, and the reply information corresponding to the first sub-question, a second sub-question closely related to the reply information corresponding to the first sub-question is obtained, improving the retrieval accuracy of the target model.

[0108] To better screen the target reply information from the candidate reply information, optionally, in the information processing method provided in the embodiments of the present application, determining the target reply information according to the first candidate reply information and at least one candidate reply information includes:

[0109] First step, calculate the evaluation score corresponding to the first candidate reply information and the evaluation scores corresponding to at least one candidate reply information according to a pre-designed calculation mechanism.

[0110] Optionally, for the obtained first candidate reply information and at least one candidate reply information, a consistency verification method can be designed to select a suitable answer. Jaccard similarity and text embedding are methods used in natural language processing to measure the similarity between texts, and these methods can be used to implement the consistency verification of text answers and the scores of each answer. Jaccard similarity mainly focuses on the overlap at the lexical level, and its value ranges from 0 to 1. The closer the value is to 1, the higher the similarity between the texts. The text embedding score focuses on semantic similarity, and similarly its value ranges from 0 to 1. The closer the value is to 1, the higher the semantic similarity of the text. The calculation formula of Jaccard similarity can be as follows:

[0111]

[0112] where N is the number of valid answer nodes, is the set at the word level of candidate reply information i, is the set at the word level of candidate reply information j.

[0113] The text embedding calculation formula can be as follows:

[0114]

[0115] where N is the number of valid answer nodes, is the embedding vector of candidate reply information i, is the embedding vector of candidate reply information j.

[0116] For example, the calculated evaluation score corresponding to the first candidate reply information is 0.8, and the evaluation score corresponding to the second candidate reply information is 0.7.

[0117] In the second step, use the response information corresponding to the highest evaluation score among the evaluation scores of the first candidate response information and the evaluation scores of at least one candidate response information as the target response information.

[0118] For example, if the evaluation score of the first candidate response information in the first step is the highest, then use the first candidate response information as the target response information.

[0119] In summary, through the above steps, it is possible to more intelligently screen candidate response information and ensure that the finally output target response information better meets the user's needs.

[0120] To better determine the target response information based on multiple candidate response information, optionally, in the information processing method provided in the embodiments of the present application, determining the target response information according to the first candidate response information and at least one candidate response information includes:

[0121] In the first step, preprocess the first candidate response information to obtain the preprocessed first candidate response information.

[0122] Optionally, the above preprocessing methods may include operations such as text cleaning, format unification, keyword extraction, etc., to ensure that the first candidate response information reaches a consistent format and quality before processing.

[0123] In the second step, preprocess at least one candidate response information to obtain the preprocessed at least one candidate response information.

[0124] Optionally, the same preprocessing operations as in the first step can be performed on at least one candidate response information to ensure that all candidate response information remains consistent in format, etc. in subsequent fusion processing.

[0125] In the third step, extract the first target information of the preprocessed first candidate response information.

[0126] Optionally, the above target information refers to the key facts, concepts, or data points extracted from the preprocessed candidate response information. For example, from the preprocessed first candidate response information, "The Yangtze River has the largest drainage area; among the tributaries of the Yangtze River, 49 tributaries have a drainage area of more than 10,000 square kilometers." can be extracted as the first target information.

[0127] In the fourth step, extract multiple target information of the preprocessed at least one candidate response information.

[0128] For example, the target information "The specific river names included in the tributaries of the Yangtze River; 49 tributaries have a drainage area of more than 10,000 square kilometers." extracted from the preprocessed second candidate response information.

[0129] In the fifth step, based on a preset fusion strategy, the first target information and multiple pieces of target information are fused to obtain the target response information.

[0130] Optionally, the fusion strategies for fusing the first target information and multiple pieces of target information may include weight assignment, content complementarity, information summarization, etc. The purpose is to combine the advantages of all target information to generate a comprehensive response, that is, the target response information. For example, it can be "Among the rivers originating from the Qinghai-Tibet Plateau, the Yangtze River ranks first with a drainage area of approximately 1.8 million square kilometers. The Yangtze River has more than 700 tributaries, including approximately 100 first-level tributaries, specifically including the Yalong River, Min River, Jialing River, Wu River, Xiang River, Gan River, etc. It should be noted that 49 tributaries of the Yangtze River have a drainage area exceeding 10,000 square kilometers, and these tributaries have an important impact on the water volume and water quality of the Yangtze River."

[0131] In summary, obtaining the target response information through the above steps ensures that the target response information covers the key information in all candidate response information, avoids information omission, and improves the comprehensiveness and richness of the response.

[0132] Optionally, Figure 4 is a schematic diagram of a method for information processing provided according to an embodiment of the present application, as Figure 4 shown, the main steps are as follows:

[0133] First, directly retrieving the question information can obtain the first candidate response information;

[0134] Next, the question information can be subject to question planning to obtain at least one question decomposition strategy corresponding to the question information, and the question information is decomposed according to various question decomposition strategies to obtain at least one question set;

[0135] In the case of obtaining a question set, this question set can be directly retrieved, and then the response information is summarized to obtain the final target response information;

[0136] In the case where a sub-question is included in the question set, the sub-question can be directly retrieved, and then the response information is summarized to obtain the target response information;

[0137] In the case of obtaining multiple question sets, according to the logical relationship between multiple sub-questions in the question sets, the next sub-question can be rewritten based on the response information of the previous sub-question, and the step of rewriting the next sub-question according to the response information of the rewritten sub-question is repeatedly executed until a second preset condition is reached, and all response information can be summarized to obtain the target response information.

[0138] In summary, after the above example performs problem planning on the problem information to obtain at least one problem set, it retrieves the problem set and summarizes the reply information to obtain the target reply information, making the target reply information more comprehensive and accurate.

[0139] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0140] Embodiment 2

[0141] The embodiment of the present application also provides an information processing device. It should be noted that the information processing device in the embodiment of the present application can be used to execute the information processing method provided in the embodiment of the present application. The following introduces the information processing device provided in the embodiment of the present application.

[0142] According to the embodiment of the present application, there is also provided a device for implementing the above information processing method, as Figure 5 shown, the device includes: a receiving unit 501, an input unit 502, a splitting unit 503, and a retrieval unit 504.

[0143] Specifically, the receiving unit 501 is configured to receive the problem information to be replied;

[0144] The input unit 502 is configured to input the problem information into the target model, and the target model performs problem planning on the problem information based on the decision retrieval algorithm to obtain various problem decomposition strategies corresponding to the problem information;

[0145] The splitting unit 503 is configured to split the problem information into at least one problem set according to various problem decomposition strategies, where there is a dependency relationship between the sub-problems in the problem set of the at least one problem set;

[0146] The retrieval unit 504 is configured to perform retrieval according to the dependency relationship between the sub-problems in the at least one problem set to obtain the target reply information corresponding to the problem information.

[0147] The information processing device provided by the embodiment of the present application receives the question information to be replied through the receiving unit 501; the input unit 502 inputs the question information into the target model, and the target model performs question planning on the question information based on the decision retrieval algorithm to obtain multiple question decomposition strategies corresponding to the question information; the splitting unit 503 splits the question information into at least one question set according to the multiple question decomposition strategies, wherein there is a dependency relationship between the sub-questions in the question set of the at least one question set; the retrieval unit 504 performs retrieval according to the dependency relationship between the sub-questions in the at least one question set to obtain the target reply information corresponding to the question information, solving the problem in the related art that when using the target model to retrieve the question information, it is easily affected by intermediate error steps, resulting in low accuracy of the reply information, and thus achieving the effect of improving the accuracy of information processing of the target model.

[0148] Optionally, the device further includes: a generating unit, configured to generate a first candidate reply information through the target model according to the question information before splitting the question information into at least one question set according to the multiple question decomposition strategies.

[0149] Optionally, the retrieval unit 504 includes: a retrieval subunit, configured to perform retrieval according to the dependency relationship between the sub-questions in the question set of the at least one question set, and summarize according to the generating unit to obtain at least one candidate reply information; a determining subunit, configured to determine the target reply information according to the first candidate reply information and the at least one candidate reply information.

[0150] Optionally, the retrieval subunit includes: a first determining module, configured to determine a prompt template when the at least one question set includes a first question set, wherein the prompt template includes prompt words and prompt examples; a first retrieval module, configured to retrieve the first question set in the at least one question set according to the prompt template to obtain a second candidate reply information; a second retrieval module, configured to repeatedly execute the step of retrieving the question set in the at least one question set that has not been retrieved according to the prompt template until a first preset condition is reached to obtain at least one candidate reply information.

[0151] Optionally, the first retrieval module includes: a first retrieval sub-module, configured to retrieve a first sub-question in the first question set according to the retrieval prompt words in the prompt template, and obtain a reply message corresponding to the first sub-question; a first determination sub-module, configured to, when the number of sub-questions in the first question set is one, use the reply message corresponding to the first sub-question as the second candidate reply message; a first rewriting sub-module, configured to, when the number of sub-questions in the first question set is multiple, rewrite the second sub-question in the first question set according to the reply message corresponding to the first sub-question and the prompt template, to obtain a rewritten second sub-question; a second retrieval sub-module, configured to retrieve the rewritten second sub-question according to the prompt template, and obtain a reply message corresponding to the rewritten second sub-question; a second rewriting sub-module, configured to repeatedly execute the steps of rewriting the third sub-question in the first question set according to the reply message corresponding to the second sub-question to obtain a reply message corresponding to the rewritten third sub-question, and rewriting the next sub-question according to the reply message corresponding to the rewritten third sub-question to obtain a reply message corresponding to the rewritten next sub-question, until a second preset condition is reached, to obtain reply messages corresponding to multiple rewritten sub-questions; a second determination sub-module, configured to analyze and process the first sub-question, the reply message corresponding to the first sub-question, multiple rewritten sub-questions, and the reply messages corresponding to the multiple rewritten sub-questions according to the prompt template, to obtain a second candidate reply message.

[0152] Optionally, the first rewriting sub-module includes: an analysis component, configured to analyze the rewriting prompt words in the prompt template and determine a rewriting task corresponding to the rewriting prompt words; an extraction component, configured to extract rewriting learning features of the rewriting example in the prompt template; a rewriting component, configured to rewrite the second sub-question based on the reply message corresponding to the first sub-question, the rewriting task, and the rewriting learning features, to obtain a rewritten second sub-question.

[0153] Optionally, the determination sub-unit includes: a calculation module, configured to calculate an evaluation score corresponding to the first candidate reply message and evaluation scores corresponding to at least one candidate reply message according to a preset calculation mechanism; a second determination module, configured to use the reply message corresponding to the highest evaluation score among the evaluation score corresponding to the first candidate reply message and the evaluation scores corresponding to the at least one candidate reply message as the target reply message.

[0154] Optionally, the determination subunit further includes: a first preprocessing module for preprocessing the first candidate response information to obtain preprocessed first candidate response information; a second preprocessing module for preprocessing at least one candidate response information to obtain preprocessed at least one candidate response information; a first extraction module for extracting first target information of the preprocessed first candidate response information; a second extraction module for extracting multiple target information of the preprocessed at least one candidate response information; and a fusion module for performing a fusion process on the first target information and the multiple target information based on a preset fusion strategy to obtain a target response information.

[0155] It should be noted here that the above receiving unit 501, input unit 502, splitting unit 503, and retrieval unit 504 correspond to steps S201 to S204 in Embodiment 1. The instances and application scenarios implemented by the four units and the corresponding steps are the same, but are not limited to the content disclosed in the above Embodiment 1. It should be noted that the above modules or units may be hardware components or software components stored in a memory (for example, memory 104) and processed by one or more processors (for example, processors 102a, 102b,..., 102n). The above modules may also be part of a device and may run in the computer terminal 10 provided in Embodiment 1.

[0156] Embodiment 3

[0157] An embodiment of the present application may provide a computer terminal, and the computer terminal may be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the above computer terminal may also be replaced with a mobile terminal or a terminal device such as an electronic device.

[0158] Optionally, in this embodiment, the above computer terminal may be located in at least one of multiple network devices in a computer network.

[0159] In this embodiment, the above computer terminal may execute program codes of the following steps in the information processing method: receiving question information to be replied; inputting the question information into a target model, and performing question planning on the question information by the target model based on a decision retrieval algorithm to obtain multiple question decomposition strategies corresponding to the question information; splitting the question information into at least one question set according to the multiple question decomposition strategies, where there is a dependency relationship between sub-questions in the question set of the at least one question set; and performing retrieval according to the dependency relationship between sub-questions in the at least one question set to obtain target response information corresponding to the question information.

[0160] Optionally, the above computer terminal may execute the program code of the following steps in the information processing method: Before splitting the problem information into at least one problem set according to multiple problem splitting strategies, the method includes: generating first candidate response information according to the problem information through a target model.

[0161] Optionally, the above computer terminal may execute the program code of the following steps in the information processing method: Retrieving according to the dependency relationship between sub-problems in the problem set to obtain the target response information corresponding to the problem information includes: Retrieving according to the dependency relationship between sub-problems of the problem sets in at least one problem set to obtain at least one candidate response information; determining the target response information according to the first candidate response information and at least one candidate response information.

[0162] Optionally, the above computer terminal may execute the program code of the following steps in the information processing method: The at least one problem set includes a first problem set. Retrieving according to the dependency relationship between sub-problems of the problem sets in at least one problem set to obtain at least one candidate response information includes: determining a prompt template, where the prompt template includes prompt words and prompt examples; retrieving the first problem set in at least one problem set according to the prompt template to obtain second candidate response information; repeatedly executing the step of retrieving the problem sets in at least one problem set that have not been retrieved according to the prompt template until a first preset condition is met to obtain at least one candidate response information.

[0163] Optionally, the above computer terminal may execute the program code of the following steps in the information processing method: The first problem set includes a first sub-problem. Retrieving the first problem set in at least one problem set according to the prompt template, the obtained second candidate reply information includes: Retrieving the first sub-problem in the first problem set according to the retrieval prompt words in the prompt template to obtain the reply information corresponding to the first sub-problem; When the number of sub-problems in the first problem set is one, using the reply information corresponding to the first sub-problem as the second candidate reply information; When the number of sub-problems in the first problem set is multiple, rewriting the second sub-problem in the first problem set according to the reply information corresponding to the first sub-problem and the prompt template to obtain the rewritten second sub-problem; Retrieving the rewritten second sub-problem according to the prompt template to obtain the reply information corresponding to the rewritten second sub-problem; Repeating the steps of rewriting the third sub-problem in the first problem set according to the reply information corresponding to the second sub-problem to obtain the reply information of the rewritten third sub-problem, and rewriting the next sub-problem according to the reply information of the rewritten third sub-problem to obtain the reply information of the rewritten next sub-problem until the second preset condition is reached to obtain the reply information corresponding to multiple rewritten sub-problems; Analyzing and processing the first sub-problem, the reply information corresponding to the first sub-problem, multiple rewritten sub-problems and the reply information corresponding to multiple rewritten sub-problems according to the prompt template to obtain the second candidate reply information.

[0164] Optionally, the above computer terminal may execute the program code of the following steps in the information processing method: Rewriting the second sub-problem in the first problem set according to the reply information corresponding to the first sub-problem and the prompt template to obtain the rewritten second sub-problem includes: Analyzing the rewriting prompt words in the prompt template to determine the rewriting task corresponding to the rewriting prompt words; Extracting the rewriting learning features of the rewritten examples in the prompt template; Based on the reply information corresponding to the first sub-problem, the rewriting task, and the rewriting learning features, rewriting the second sub-problem to obtain the rewritten second sub-problem.

[0165] Optionally, the above computer terminal may execute the program code of the following steps in the information processing method: Determining the target reply information according to the first candidate reply information and at least one candidate reply information includes: Calculating the evaluation score corresponding to the first candidate reply information and the evaluation scores corresponding to at least one candidate reply information according to the preset calculation mechanism; Using the reply information corresponding to the highest evaluation score among the evaluation score corresponding to the first candidate reply information and the evaluation scores corresponding to at least one candidate reply information as the target reply information.

[0166] Optionally, the above computer terminal may execute program code for the following steps in the information processing method: determining a target response message based on a first candidate response message and at least one candidate response message, including: preprocessing the first candidate response message to obtain a preprocessed first candidate response message; preprocessing the at least one candidate response message to obtain preprocessed at least one candidate response message; extracting first target information of the preprocessed first candidate response message; extracting multiple target information of the preprocessed at least one candidate response message; and based on a preset fusion strategy, performing a fusion process on the first target information and the multiple target information to obtain a target response message.

[0167] Optionally, Figure 6 is a structural block diagram of an electronic device according to an embodiment of the present application. As Figure 6 shown, the electronic device may include: one or more ( Figure 6 only one is shown in the figure) processors 602, a memory 604, a storage controller, and a peripheral interface, where the peripheral interface is connected to a radio frequency module, an audio module, and a display.

[0168] Among them, the memory may be used to store software programs and modules, such as program instructions / modules corresponding to the information processing method and device in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, implementing the above information processing method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory may further include a memory remotely disposed relative to the processor, and these remote memories may be connected to the terminal through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0169] The processor may call the information and application programs stored in the memory through a transmission device to execute the above steps in the above information processing method.

[0170] By adopting the embodiment of the present application, a solution for information processing is provided. By receiving the question information to be replied; inputting the question information into the target model, and performing question planning on the question information by the target model based on the decision retrieval algorithm to obtain multiple question decomposition strategies corresponding to the question information; splitting the question information into at least one question set according to the multiple question decomposition strategies, wherein there is a dependency relationship between the sub-questions in the question set of the at least one question set; retrieving according to the dependency relationship between the sub-questions in the at least one question set to obtain the target reply information corresponding to the question information, thereby achieving the purpose of improving the accuracy of information processing of the target model, and further solving the technical problem that when using the target model to retrieve the question information, it is easily affected by intermediate error steps, resulting in a low accuracy rate of the reply information.

[0171] Those of ordinary skill in the art can understand that Figure 6 The structure shown is only for illustration, and the electronic device can also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a palm computer, and a mobile Internet device (Mobile Internet Devices, MID), a PAD and other terminal devices. Figure 6 It does not limit the structure of the above-mentioned electronic device. For example, the electronic device may further include more or fewer components (such as a network interface, a display device, etc.) than those shown in Figure 6 or have a different configuration from that shown in Figure 6 shown.

[0172] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware of the terminal device through a program, and the program can be stored in a computer-readable storage medium. The storage medium may include: a flash drive, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disc, etc.

[0173] Embodiment 4

[0174] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the above storage medium can be used to save the program code executed by the information processing method provided in the first embodiment above.

[0175] Optionally, in this embodiment, the above storage medium can be located in any one of the computer terminals in the computer terminal group in the computer network, or in any one of the mobile terminals in the mobile terminal group.

[0176] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: receiving question information to be replied; inputting the question information into a target model, and performing question planning on the question information by the target model based on a decision retrieval algorithm to obtain multiple question decomposition strategies corresponding to the question information; splitting the question information into at least one question set according to the multiple question decomposition strategies, wherein there is a dependency relationship between sub-questions in the question set of the at least one question set; and performing retrieval according to the dependency relationship between sub-questions in the at least one question set to obtain target reply information corresponding to the question information.

[0177] Optionally, the storage medium is further configured to store program code for performing the following steps: before splitting the question information into at least one question set according to the multiple question decomposition strategies, the method includes: generating a first candidate reply information by the target model according to the question information.

[0178] Optionally, the storage medium is further configured to store program code for performing the following steps: performing retrieval according to the dependency relationship between sub-questions in the question set to obtain target reply information corresponding to the question information includes: performing retrieval according to the dependency relationship between sub-questions in the question set of the at least one question set to obtain at least one candidate reply information; and determining the target reply information according to the first candidate reply information and the at least one candidate reply information.

[0179] Optionally, the storage medium is further configured to store program code for performing the following steps: the at least one question set includes a first question set, and performing retrieval according to the dependency relationship between sub-questions in the question set of the at least one question set to obtain at least one candidate reply information includes: determining a prompt template, where the prompt template includes prompt words and prompt examples; performing retrieval on the first question set in the at least one question set according to the prompt template to obtain a second candidate reply information; and repeating the step of performing retrieval on the question sets in the at least one question set that have not been retrieved according to the prompt template until a first preset condition is met to obtain at least one candidate reply information.

[0180] Optionally, the storage medium is further configured to store program code for performing the following steps: The first question set includes a first sub-question. Retrieving the first question set in at least one question set according to a prompt template, the obtained second candidate reply information includes: Retrieving the first sub-question in the first question set according to the retrieval prompt words in the prompt template to obtain the reply information corresponding to the first sub-question; When the number of sub-questions in the first question set is one, using the reply information corresponding to the first sub-question as the second candidate reply information; When the number of sub-questions in the first question set is multiple, rewriting the second sub-question in the first question set according to the reply information corresponding to the first sub-question and the prompt template to obtain the rewritten second sub-question; Retrieving the rewritten second sub-question according to the prompt template to obtain the reply information corresponding to the rewritten second sub-question; Repeating the step of rewriting the third sub-question in the first question set according to the reply information corresponding to the second sub-question to obtain the reply information corresponding to the rewritten third sub-question, and rewriting the next sub-question according to the reply information corresponding to the rewritten third sub-question to obtain the reply information corresponding to the rewritten next sub-question until a second preset condition is reached to obtain the reply information corresponding to multiple rewritten sub-questions; Analyzing and processing the first sub-question, the reply information corresponding to the first sub-question, multiple rewritten sub-questions and the reply information corresponding to multiple rewritten sub-questions according to the prompt template to obtain the second candidate reply information.

[0181] Optionally, the storage medium is further configured to store program code for performing the following steps: Rewriting the second sub-question in the first question set according to the reply information corresponding to the first sub-question and the prompt template, the obtained rewritten second sub-question includes: Analyzing the rewriting prompt words in the prompt template to determine the rewriting task corresponding to the rewriting prompt words; Extracting the rewriting learning features of the rewriting examples in the prompt template; Based on the reply information corresponding to the first sub-question, the rewriting task and the rewriting learning features, rewriting the second sub-question to obtain the rewritten second sub-question.

[0182] Optionally, the storage medium is further configured to store program code for performing the following steps: Determining the target reply information according to the first candidate reply information and at least one candidate reply information includes: Calculating the evaluation score corresponding to the first candidate reply information and the evaluation scores corresponding to at least one candidate reply information according to a preset calculation mechanism; Using the reply information corresponding to the highest evaluation score among the evaluation score corresponding to the first candidate reply information and the evaluation scores corresponding to at least one candidate reply information as the target reply information.

[0183] Optionally, the storage medium is further configured to store program code for performing the following steps: determining a target response message based on the first candidate response message and at least one candidate response message, including: preprocessing the first candidate response message to obtain a preprocessed first candidate response message; preprocessing the at least one candidate response message to obtain preprocessed at least one candidate response message; extracting first target information from the preprocessed first candidate response message; extracting multiple target information from the preprocessed at least one candidate response message; and performing a fusion process on the first target information and the multiple target information based on a preset fusion strategy to obtain the target response message.

[0184] The present application also provides a computer program product, which is adapted to execute a program of the information processing method steps when executed on a data processing device.

[0185] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0186] In the above embodiments of the present application, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0187] In several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0188] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0189] In addition, each functional unit in the various embodiments of the present application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0190] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical discs.

[0191] The above are only the preferred embodiments of this application. It should be noted that for those of ordinary skill in the art, without departing from the principle of this application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this application.

Claims

1. An information processing method, characterized in that: include: Receive information about questions to be answered; Inputting the problem information into the target model, and performing problem planning on the problem information based on the decision retrieval algorithm through the target model to obtain a plurality of problem solving strategies corresponding to the problem information; Splitting the problem information into at least one problem set according to the multiple problem decomposition strategies, wherein there is a dependency relationship between sub-problems in the problem set of the at least one problem set; Search is performed based on the dependency relationship between the sub-questions in the at least one question set to obtain target response information corresponding to the question information.

2. The method according to claim 1, characterized in that Before splitting the problem information into at least one problem set according to the multiple problem splitting strategies, the method includes: Generate first candidate reply information based on the question information through the target model.

3. The method according to claim 2, characterized in that The target reply information corresponding to the question information is retrieved according to the dependency relationship between the sub-questions in the question set, and includes: Retrieving according to the dependency relationship between the sub-questions of the question set in the at least one question set to obtain at least one candidate answer information; The target reply information is determined according to the first candidate reply information and the at least one candidate reply information.

4. The method according to claim 3, characterized in that The at least one question set includes a first question set, and searching based on the dependency relationship between sub-questions of the question set in the at least one question set to obtain at least one candidate reply information includes: Determine a prompt template, wherein the prompt template includes a prompt word and a prompt sample; Retrieving a first question set in the at least one question set according to the prompt template to obtain second candidate answer information; The step of searching for a question set that has not been searched in the at least one question set according to the prompt template is repeatedly performed until a first preset condition is met, thereby obtaining the at least one candidate reply information.

5. The method according to claim 4, characterized in that The first question set includes a first sub-question, and the first question set in the at least one question set is retrieved according to the prompt template to obtain the second candidate reply information including: Searching for the first sub-question in the first question set according to the search prompt word in the prompt template to obtain answer information corresponding to the first sub-question; When the number of sub-questions in the first question set is one, taking the answer information corresponding to the first sub-question as the second candidate answer information; In the case where there are multiple sub-questions in the first question set, rewriting the second sub-question in the first question set according to the reply information corresponding to the first sub-question and the prompt template to obtain a rewritten second sub-question; The rewritten second sub-question is searched according to the prompt template to obtain the reply information corresponding to the rewritten second sub-question; Repeat the steps of rewriting the third sub-question in the first question set according to the reply information corresponding to the second sub-question to obtain the reply information of the rewritten third sub-question, and rewriting the next sub-question according to the rewritten reply information of the third sub-question to obtain the rewritten reply information of the next sub-question, until the second preset condition is met and reply information corresponding to multiple rewritten sub-questions is obtained; The first sub-question, the reply information corresponding to the first sub-question, the multiple rewritten sub-questions, and the reply information corresponding to the multiple rewritten sub-questions are analyzed and processed according to the prompt template to obtain second candidate reply information.

6. The method according to claim 5, characterized in that The second sub-question in the first question set is rewritten according to the reply information corresponding to the first sub-question and the prompt template, and the rewritten second sub-question includes: Analyzing the rewriting prompt words in the prompt template to determine the rewriting tasks corresponding to the rewriting prompt words; Extracting rewriting learning features of the rewriting examples in the prompt template; The second sub-problem is rewritten based on the reply information corresponding to the first sub-problem, the rewriting task, and the rewriting learning feature to obtain a rewritten second sub-problem.

7. The method according to claim 3, characterized in that Determining target reply information according to the first candidate reply information and the at least one candidate reply information includes: Calculate the evaluation score corresponding to the first candidate reply information and the evaluation score corresponding to the at least one candidate reply information according to a preset calculation mechanism; The reply information corresponding to the highest evaluation score between the evaluation score corresponding to the first candidate reply information and the evaluation score corresponding to the at least one candidate reply information is used as the target reply information.

8. The method according to claim 3, characterized in that Determining target reply information according to the first candidate reply information and the at least one candidate reply information includes: Preprocessing the first candidate reply information to obtain preprocessed first candidate reply information; Preprocessing the at least one candidate reply information to obtain at least one preprocessed candidate reply information; Extracting first target information of the preprocessed first candidate reply information; extracting a plurality of target information of the at least one candidate reply information after the preprocessing; Based on a preset fusion strategy, the first target information and the multiple target information are fused to obtain target reply information.

9. An information processing device, characterized in that: include: A receiving unit, used for receiving question information to be answered; An input unit, used to input the problem information into a target model, and perform problem planning on the problem information based on a decision retrieval algorithm through the target model to obtain a plurality of problem solving strategies corresponding to the problem information; A splitting unit, configured to split the problem information into at least one problem set according to the multiple problem splitting strategies, wherein there is a dependency relationship between sub-problems in the problem set of the at least one problem set; The retrieval unit is used to perform retrieval based on the dependency relationship between the sub-questions in the at least one question set to obtain target response information corresponding to the question information.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the computer-readable storage medium is located is controlled to execute the information processing method according to any one of claims 1 to 8.

11. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program, when running, executes the information processing method according to any one of claims 1 to 8.

12. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the information processing method according to any one of claims 1 to 8 are implemented.

Images