Method and device for processing multi-document question and answer task, electronic equipment and storage medium
By fine-tuning the Q&A big model and sorting candidate documents, the problem of insufficient data in the traditional Q&A big model in multi-document Q&A tasks is solved, and the processing efficiency and accuracy are improved, especially when processing long texts and complex data such as scientific research articles.
Patent Information
- Application Number
- CN202510196814.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-21
- Publication Date
- 2025-06-10
AI Technical Summary
When handling multi-document Q&A tasks, traditional Q&A models have insufficient processing efficiency and accuracy due to insufficient data volume, especially when dealing with areas with strong professionalism or rapid updates, their knowledge may not be comprehensive or timely.
By fine-tuning the Q&A big model with training data for multi-document Q&A tasks, training data for scientific research articles is generated, candidate documents are reordered through vector search and neural network, and finally, the fine-tuned Q&A big model is used to generate answers.
Improve the ability of Q&A big model to process long text and complex data, and enhance the performance of processing efficiency and accuracy in multi-document Q&A tasks.
Smart Images

Figure CN120123476A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of natural language processing, and in particular, to a method, apparatus, electronic device, and storage medium for processing multi-document question answering tasks. Background Art
[0002] Multi-Document Question Answering (MDQA) is a method of using information from multiple documents to answer user questions. MDQA can integrate information from multiple source documents to provide more comprehensive and accurate answers to user questions. Summary of the Invention
[0003] Question answering large models in the related art (such as generative language models) can handle multi-document question answering tasks to a certain extent, but their processing efficiency and accuracy are insufficient in dealing with unstructured data, coping with data insufficiency or information dispersion. For example, when dealing with fields with strong professionalism or rapid updates, some question answering large models may rely on knowledge that is not comprehensive or timely enough.
[0004] In view of this, the present disclosure provides a method, apparatus, electronic device, and storage medium for processing multi-document question answering tasks, which solves the problem of insufficient processing efficiency and accuracy of traditional question answering large models in processing multi-document question answering tasks due to insufficient data volume, and improves the ability of the question answering large model to process long texts and complex data such as scientific research articles.
[0005] In a first aspect, an embodiment of the present disclosure provides a method for processing multi-document question answering tasks, including:
[0006] Fine-tuning a question answering large model using training data for multi-document question answering tasks;
[0007] Obtaining an initial retrieval result for a user question using the fine-tuned question answering large model;
[0008] Selecting a set of candidate documents from the initial retrieval result and re-ranking the candidate documents; and
[0009] Generating an answer to the user question based on the re-ranked candidate documents using the question answering large model.
[0010] According to an embodiment of the present disclosure, the training data is generated based on scientific research articles, and the training data is generated by a large language model through the following operations:
[0011] Extracting key paragraphs from scientific research articles according to prompts; and
[0012] Generating the training data based on the key paragraphs according to prompts.
[0013] According to an embodiment of the present disclosure, selecting a set of candidate documents from the initial retrieval results and re - ranking the candidate documents includes:
[0014] Selecting a set of candidate documents from the initial retrieval results through vector retrieval;
[0015] Using a neural network to analyze the relevance between each candidate document and the user question to obtain a relevance score for each candidate document; and
[0016] Re - ranking the candidate documents according to the relevance score and a predetermined rule.
[0017] According to an embodiment of the present disclosure, the predetermined rule is designed based on the weight value of the candidate document and a sensitive word list, and the weight value of the candidate document indicates the reliability of the source of the candidate document.
[0018] In a second aspect, an embodiment of the present disclosure provides an apparatus for processing multi - document question - answering tasks, including:
[0019] A question - answering large - model fine - tuning unit configured to fine - tune a question - answering large model using training data for multi - document question - answering tasks;
[0020] An initial retrieval unit configured to obtain an initial retrieval result for a user question using the fine - tuned question - answering large model;
[0021] A candidate document re - ranking unit configured to select a set of candidate documents from the initial retrieval results and re - rank the candidate documents; and
[0022] An answer generation unit configured to generate an answer for the user question using the question - answering large model according to the re - ranked candidate documents.
[0023] According to an embodiment of the present disclosure, the training data is generated based on scientific research articles, and the apparatus further includes:
[0024] A training data generation unit configured to generate the training data using a large - language model through the following operations:
[0025] Extracting key paragraphs from scientific research articles according to prompting words; and
[0026] Generating the training data based on the key paragraphs according to prompting words.
[0027] According to an embodiment of the present disclosure, the candidate document re - ranking unit is configured to:
[0028] Select a set of candidate documents from the initial retrieval results through vector retrieval;
[0029] Analyze the relevance between each candidate document and the user question using a neural network to obtain a relevance score for each candidate document; and
[0030] Reorder the candidate documents according to the relevance score and a predetermined rule.
[0031] In a third aspect, an embodiment of the present disclosure provides an electronic device, which includes:
[0032] At least one processor; and
[0033] A memory communicatively connected to the at least one processor, wherein
[0034] The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method for processing multi-document question answering tasks according to the embodiments of the present disclosure.
[0035] In a fourth aspect, an embodiment of the present disclosure provides a non-transitory computer-readable storage medium, which stores computer instructions for causing a computer to execute the method for processing multi-document question answering tasks according to the embodiments of the present disclosure.
[0036] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, and when the program instructions are executed by a computer, the computer is enabled to execute the method for processing multi-document question answering tasks according to the embodiments of the present disclosure.
[0037] The method, apparatus, electronic device, and storage medium for processing multi-document question answering tasks provided by the present disclosure solve the problems of insufficient processing efficiency and accuracy of traditional question answering large models when processing multi-document question answering tasks due to insufficient data volume, and improve the ability of the question answering large model to process long texts and complex data such as scientific research articles. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.
[0039] Figure 1 It is an exemplary flowchart of the method for processing multi-document question answering tasks provided by the embodiments of the present disclosure;
[0040] Figure 2Exemplary structural diagram of the apparatus for processing multi-document question answering tasks provided by embodiments of the present disclosure;
[0041] Figure 3 Shows an exemplary structural diagram of a device capable of implementing the method according to embodiments of the present disclosure. Detailed implementation manners
[0042] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.
[0043] The following uses specific specific examples to illustrate the implementation manners of the present disclosure. Those skilled in the art can easily understand other advantages and effects of the present disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without making creative efforts belong to the scope of protection of the present disclosure.
[0044] It should be noted that the following describes various aspects of embodiments within the scope of the appended claims. It should be obvious that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is illustrative only. Based on the present disclosure, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, this device can be implemented and this method can be practiced using other structures and / or functions in addition to one or more of the aspects described herein.
[0045] It should also be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present disclosure in a schematic manner. The diagrams only show the components related to the present disclosure, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in its actual implementation can be an arbitrary change, and the component layout type may also be more complex.
[0046] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.
[0047] The present disclosure provides a method for processing a multi-document question-answering task, which can be performed by an electronic device, in particular, by one or more processors in the electronic device. In the present disclosure, the electronic device can be a device with data processing capabilities such as a laptop computer, a desktop computer, a smart mobile terminal, a workstation or a server.
[0048] like Figure 1 As shown, the method 100 for processing a multi-document question answering task according to an embodiment of the present disclosure includes the following steps.
[0049] In step S101, the question-answering model is fine-tuned using the training data for the multi-document question-answering task.
[0050] In some embodiments, in order to solve the problem that the knowledge that the large question-answering model relies on when processing multi-document question-answering tasks in the field of scientific research may not be comprehensive or timely, training data for multi-document question-answering tasks can be generated based on scientific research articles. In the present disclosure, a large language model such as a generative language model is used to efficiently and conveniently generate large quantities of training data for multi-document question-answering tasks based on scientific research articles. Specifically, for example, the large language model can be used to generate training data through the following operations.
[0051] First, the large language model extracts the key paragraphs in the scientific research article according to the pre-designed appropriate prompt words. The scientific research article can be, for example, an academic journal article, a conference paper, a literature review, a graduation thesis, an online popular science article, a science and technology forum article, etc. The large language model can be, for example, a generative language model that can give the user's expected answer according to the prompt words, or other types of large language models, as long as the functions described for the large language model in this disclosure can be realized. Appropriate prompt words can be pre-designed so that the training data generated by the large language model fully reflects the core content of the scientific research article. In an example, the pre-designed prompt words can be, for example, "Given a scientific article excerpt, identify key information points and organize the minto a structured dictionary format (given a scientific article excerpt, determine the key information points and organize them into a structured dictionary format)". After the prompt words are input, the large language model conducts an in-depth analysis of the structure of the scientific research article, and uses natural language processing technology to identify and extract key paragraphs in the scientific research article, such as article abstracts, research hypotheses, experimental procedures, experimental data, data analysis results, etc.
[0052] Then, the large language model generates training data based on the key paragraphs of the extracted scientific research articles according to pre-designed appropriate prompting words. For example, training data generation tasks can be designed for the large language model for different types of scientific research article paragraphs. For example, in the case where the extracted key paragraph is the article abstract, the large language model can be made to rewrite the article abstract to generate training data; in the case where the extracted key paragraph is experimental data, the large language model can be made to simplify complex data paragraphs to generate training data. Corresponding prompting words can be designed to guide the training data generation process. For example, for the above two cases, the prompting words used can be "Rewrite the abstract to make it concise yet informative" and "Simplify a complex data paragraph while retaining key information" respectively. After being input with the prompting words, the large language model processes the key paragraphs of the extracted scientific research articles according to the prompting words to generate high-quality, diverse training data specifically for the multi-document question answering task for fine-tuning the question answering large model.
[0053] After generating the aforementioned training data, the training data can be used to fine-tune the question answering large model so that the question answering large model can perform better in the scientific research field for the multi-document question answering task. Thus, the challenges faced by traditional question answering large models in processing multi-document question answering tasks due to insufficient data volume are solved, and the ability of the question answering model to process long texts and complex data such as scientific research articles can be significantly improved.
[0054] Next, in step S102, the fine-tuned question answering large model is used to obtain an initial retrieval result for the user's question.
[0055] In some embodiments, after fine-tuning the question answering large model with the training data for the multi-document question answering task, the user can directly input the question for which an answer is expected to the question answering large model, or can input the question through, for example, the application program interface provided by the device for processing the multi-document question answering task according to the embodiments of the present disclosure. After receiving the user's question, the question answering large model retrieves for the user's question to obtain an initial retrieval result for the user's question. In the present disclosure, the initial retrieval result can include multiple documents related to the user's question.
[0056] Next, in step S103, a set of candidate documents is selected from the initial retrieval result and the candidate documents are re-ranked.
[0057] In some embodiments, a re-rank algorithm that combines a predefined rule and a neural network is used to enhance the security and accuracy in the process of processing multi-document question answering tasks. This algorithm combines the advantages of the predefined rule and the neural network, improves the ability of the question answering large model to process structured data, and provides a more accurate reference document ranking and selection for complex multi-document question answering tasks. Specifically, this algorithm may include the following operations, for example.
[0058] First, a set of candidate documents is selected from the initial retrieval results through vector retrieval. For example, Embedding models such as Transformer embedding or Text-Ada-002 can be used for vector retrieval to select a set of candidate documents with high relevance to the user's question from multiple documents in the initial retrieval results. Next, a neural network such as BGE Re-Ranker can be used to analyze the relevance (text similarity) between each candidate document and the user's question to obtain the relevance score of each candidate document. Then, considering the relevance score and the predefined rule comprehensively, the candidate documents are re-ranked. The predefined rule can be designed based on, for example, a weight value indicating the reliability of the source of the candidate document and a sensitive word list (such as laws and regulations, social norms, and culture). For example, the candidate documents can be re-ranked from high to low according to the relevance score first, and then the re-ranked order can be adjusted according to the reliability weight value and whether it involves sensitive words. Or, weights can also be set for the relevance score and the predefined rule, and the candidate documents are re-ranked according to the weighted sum of the relevance score and the predefined rule of each candidate document. In this algorithm, the neural network mainly gives the relevance score of the candidate documents according to the text similarity, while the predefined rule can not only focus on the reliability of the source of the candidate document but also care about the sensitive word list, so that the finally re-ranked candidate documents given by this algorithm are both accurate and meet the security specifications.
[0059] Next, in step S104, using the question answering large model, an answer to the user's question is generated according to the re-ranked candidate documents. Since the question answering large model is fine-tuned using the training data for multi-document question answering tasks, and the candidate documents are also re-ranked using the predefined rule and the neural network, therefore, after performing the previous steps S101 to S103, using the question answering large model to generate an answer to the user's question according to the re-ranked candidate documents can solve the problems of insufficient processing efficiency and accuracy of the question answering large model when processing multi-document question answering tasks, and improve the ability of the question answering large model to process long texts and complex data such as scientific research articles.
[0060] The embodiments of the present disclosure also provide a device for processing multi-document question answering tasks. As Figure 2As shown, the apparatus 200 for processing multi-document question answering tasks may include: a question answering large model fine-tuning unit 201, configured to fine-tune a question answering large model by using training data for multi-document question answering tasks; an initial retrieval unit 202, configured to obtain an initial retrieval result for a user question by using the fine-tuned question answering large model; a candidate document re-ranking unit 203, configured to select a set of candidate documents from the initial retrieval result and re-rank the candidate documents; and an answer generation unit 204, configured to generate an answer for the user question based on the re-ranked candidate documents by using the question answering large model.
[0061] In some embodiments, the training data is generated based on scientific research articles. In this case, although Figure 2 not shown in the figure, the apparatus 200 may further include: a training data generation unit, configured to generate training data by using a large language model through the following operations: extracting key paragraphs in a scientific research article according to a prompt; and generating training data based on the extracted key paragraphs according to the prompt.
[0062] In some embodiments, the candidate document re-ranking unit 203 is further configured to: select a set of candidate documents from the initial retrieval result through vector retrieval; analyze the relevance between each candidate document and the user question by using a neural network to obtain a relevance score for each candidate document; and re-rank the candidate documents according to the relevance score and a predetermined rule.
[0063] For the specific functions of the units of the apparatus 200, reference may be made to the foregoing description of the method for processing multi-document question answering tasks according to embodiments of the present disclosure, and details are not described herein again.
[0064] Embodiments of the present disclosure further provide a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer is caused to execute the method for processing multi-document question answering tasks according to embodiments of the present disclosure.
[0065] It should be understood that the foregoing various specific algorithms, neural networks, large language models, and question answering large models are only examples and not limitations. Those skilled in the art can use other various specific algorithms, neural networks, large language models, and question answering large models to implement the method according to embodiments of the present disclosure without departing from the spirit and scope of the present disclosure.
[0066] Figure 3FIG. shows an exemplary structural diagram of a device 1000 that can implement the method according to an embodiment of the present disclosure or implement the device according to an embodiment of the present disclosure. In some embodiments, it may include more or fewer devices than shown in the figure. In some embodiments, it may be implemented using a single or multiple devices. In some embodiments, it may be implemented using cloud or distributed devices.
[0067] As Figure 3 shown, the device 1000 includes a processor 1001, which can perform various appropriate operations and processes according to the programs and / or data stored in the read-only memory (ROM) 1002 or the programs and / or data loaded from the storage section 1008 into the random access memory (RAM) 1003. The processor 1001 can be a multi-core processor or can include multiple processors. In some embodiments, the processor 1001 can include a general main processor and one or more special coprocessors, such as a central processing unit (CPU), a graphics processing unit (GPU), a neural network processing unit (NPU), a digital signal processor (DSP), and so on. In the RAM 1003, various programs and data required for the operation of the device 1000 are also stored. The processor 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. The input / output (I / O) interface 1005 is also connected to the bus 1004.
[0068] The above-mentioned processor and memory are jointly used to execute the programs stored in the memory, and when the programs are executed by a computer, they can implement the methods, steps, or functions described in the above embodiments.
[0069] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, a touch screen, etc.; an output section 1007 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed so that the computer program read from it can be installed into the storage section 1008 as needed. Figure 3 Only some components are schematically shown, and it does not mean that the device 1000 only includes Figure 3 the components shown.
[0070] The systems, devices, modules, or units illustrated in the above embodiments can be implemented by a computer or its associated components. The computer can be, for example, a mobile terminal, a smart phone, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a personal digital assistant, a media player, a navigation device, a game console, a tablet computer, a wearable device, a smart TV, an Internet of Things system, a smart home, an industrial computer, a server, or a combination thereof.
[0071] Although not shown, in the embodiments of the present disclosure, a computer-readable storage medium is provided, on which a computer program / instructions are stored, and when the computer program / instructions are executed by a processor, the methods described in the embodiments of the present disclosure are implemented.
[0072] The storage media of the embodiments of the present disclosure include items that are permanent and non-permanent, removable and non-removable, and can implement information storage by any method or technology. Examples of storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device.
[0073] Although not shown, the embodiments of the present disclosure also provide a computer program product, including: a computer program / instructions, and when the computer program / instructions are executed by a processor, the methods described in the embodiments of the present disclosure are implemented.
[0074] The methods, programs, systems, devices, etc. of the embodiments of the present disclosure can be executed or implemented in a single or multiple networked computers, and can also be practiced in a distributed computing environment. In the embodiments, in these distributed computing environments, tasks can be executed by remote processing devices connected through a communication network.
[0075] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, those skilled in the art can think that the implementation of the functional modules / units or controllers and related method steps illustrated in the above embodiments can be achieved in a manner combining software, hardware, and soft / hardware.
[0076] Unless explicitly stated, the actions or steps of the methods and programs recorded according to the embodiments of the present disclosure do not necessarily have to be executed in a specific order and can still achieve the desired results. In some embodiments, multi-tasking and parallel processing are also possible or may be advantageous.
[0077] In this document, multiple embodiments of the present disclosure are described. For the sake of brevity, the descriptions of each embodiment are not exhaustive, and the same or similar features or parts between various embodiments may be omitted. In this document, "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean applicable to at least one embodiment or example according to the present disclosure, rather than all embodiments. The above terms do not necessarily refer to the same embodiment or example. Without contradiction, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples.
[0078] The exemplary systems and methods of the present disclosure have been specifically shown and described with reference to the above embodiments, which are only examples of the best mode for implementing the systems and methods. Those skilled in the art can understand that various changes can be made to the embodiments of the systems and methods described herein when implementing the systems and / or methods without departing from the spirit and scope of the present disclosure defined in the appended claims.
Claims
1. A method for processing multi-document question answering tasks, characterized in that: include: Fine-tune the large question-answering model using training data for multi-document question-answering tasks; Using the fine-tuned question-answering model, initial retrieval results for the user's question are obtained; Selecting a set of candidate documents from the initial search results and re-ranking the candidate documents; as well as The question-answering model is used to generate answers to the user's questions based on the re-ranked candidate documents.
2. The method for processing multi-document question answering tasks according to claim 1, characterized in that: The training data is generated based on scientific research articles and is generated using a large language model through the following operations: Extract key paragraphs from scientific research articles based on prompt words; and According to the prompt words, the training data is generated based on the key paragraphs.
3. The method for processing multi-document question answering tasks according to claim 1, characterized in that: Selecting a set of candidate documents from the initial search results and re-ranking the candidate documents includes: Selecting a set of candidate documents from the initial search results by vector retrieval; Analyzing the relevance of each candidate document to the user question using a neural network to obtain a relevance score for each candidate document; and The candidate documents are reordered according to the relevance score and a predetermined rule.
4. The method for processing multi-document question answering tasks according to claim 3, characterized in that: The predetermined rule is designed based on a weight value of a candidate document and a sensitive word list, and the weight value of the candidate document indicates the reliability of a source of the candidate document.
5. A device for processing multi-document question answering tasks, characterized in that: include: A question-answering large model fine-tuning unit is configured to fine-tune the question-answering large model using training data for a multi-document question-answering task; An initial retrieval unit is configured to obtain initial retrieval results for user questions using the fine-tuned question-answering model; a candidate document re-ranking unit, configured to select a group of candidate documents from the initial search results and re-rank the candidate documents; as well as The answer generation unit is configured to use the question-answering model to generate an answer to the user's question based on the re-ranked candidate documents.
6. The device for processing multi-document question answering tasks according to claim 5, characterized in that: The training data is generated based on scientific research articles, and the device also includes: The training data generating unit is configured to generate the training data by using the large language model through the following operations: Extract key paragraphs from scientific research articles based on prompt words; and According to the prompt words, the training data is generated based on the key paragraphs.
7. The device for processing multi-document question-answering tasks according to claim 5, characterized in that: The candidate document rearrangement unit is configured to: Selecting a set of candidate documents from the initial search results by vector retrieval; Analyzing the relevance of each candidate document to the user question using a neural network to obtain a relevance score for each candidate document; as well as The candidate documents are reordered according to the relevance score and a predetermined rule.
8. An electronic device, characterized in that: The electronic device includes: at least one processor; and a memory communicatively coupled to the at least one processor, wherein: The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for processing a multi-document question and answer task according to any one of claims 1 to 4.
9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, which are used to enable a computer to execute the method for processing a multi-document question-answering task according to any one of claims 1 to 4.
10. A computer program product, comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, which, when executed by a computer, cause the computer to execute the method for processing a multi-document question-answering task according to any one of claims 1 to 4.
