Project management experience processing method based on RAG and storage medium

Abstract

The invention discloses an RAG-based project management experience processing method and a storage medium, and the method combines two functions of retrieval and generation of an RAG technology, firstly retrieves related information, then generates a high-quality natural language text, and finally integrates the high-quality natural language text into a question and answer system, and the project management experience processing efficiency is improved through an intelligent chat mode. An accurate technical scheme can be quickly provided for a project manager, and the project manager is helped to efficiently identify and solve potential problems in a project, so that the decision quality and execution efficiency of project management are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and machine learning, and in particular to a RAG-based project management experience processing method and storage medium. Background Art

[0002] The lessons learned during project management are of great value in improving project success rates and optimizing decision support. However, traditional experience summary and sharing methods mainly rely on manual analysis and report writing, which have problems such as long information processing cycle, low data analysis accuracy, and poor information sharing efficiency.

[0003] To obtain the experience and lessons of project management, it is generally necessary to manually search and process them. Manual search and processing has the defects of low efficiency and inability to quickly find the required experience and lessons. With the development of digital technology, the project management process has also developed from traditional paper management to digital data management. Although this management method allows users to search and obtain data, it still does not get rid of the user search operation. User search will lead to inaccurate search time, search results, etc. due to different users' search skills, affecting the progress of project development, and unable to accurately and reliably provide guidance to current project users based on project development experience and lessons. How to use these project development experience and lessons is very important for project managers. Existing technology cannot solve the problem of accurate and rapid use of project experience and lessons. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a RAG-based project management experience processing method and storage medium, which can efficiently extract and analyze the lessons learned in the project management process through automated means, and provide the results to project managers in a real-time and accurate manner, thereby significantly improving the decision-making quality and execution efficiency of project management.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: a project management experience processing method and storage medium based on RAG, firstly, the original data is trained to obtain summary and analysis text. Secondly, the lessons learned are summarized and integrated into the Dify chat tool. Finally, regular updates are performed to ensure the accuracy and timeliness of decision-making. The specific steps include:

[0006] S1. Identify and extract various data involved in the project management process from relevant data sources and import them into the system.

[0007] S2: Segment the data imported into the system, then vectorize the segmented data and store the data by building an efficient index mechanism.

[0008] S3. Use RAG technology to analyze data. In the retrieval stage, first use the preset query conditions to retrieve information related to the current project from the stored database. In the generation stage, the retrieved relevant data is input into the pre-trained language model, and the information is analyzed and understood by the model to generate a summary and analysis text in natural language.

[0009] S4. By using preset templates or structured methods, the generated summary text is structured to extract the key lessons learned in the project management process to facilitate user understanding.

[0010] S5. Integrate the entire process into an automated tool. By integrating the Dify chat tool, the lessons learned can be easily obtained through chat questions and answers instead of relying on precise keyword searches, which reduces the difficulty of obtaining lessons learned and increases the breadth of lessons learned.

[0011] S6. Regular update mechanism to automatically extract the latest data from relevant data sources and generate new lessons learned.

[0012] The step S1 specifically comprises:

[0013] S11 loads data through the REST API interface of the relevant data source, and selects the appropriate format according to the needs, such as obtaining project, task, user and other information in JSON format. Due to the differences in different data sources and formats, all data must be converted into a unified paradigm when loading data.

[0014] S12 filters, cleans and repairs the loaded data, removes redundant or unnecessary data, and fills or deletes incomplete data. At the same time, it compresses the loaded data to increase the data analysis speed and save memory space, and imports the processed data into the target system.

[0015] The step S2 specifically comprises:

[0016] After reading the processed data from the system, S21 uses natural language processing (NLP) technology to accurately segment the data, divide the text into sentences or words, and store these segmented fragments in a list in an orderly manner.

[0017] S22 takes the segmented data as input, processes it using vectorization technology, converts it into a digital vector form, and generates a corresponding vector matrix.

[0018] S23 takes the vector matrix as input and uses the approximate nearest neighbor (ANN) algorithm to quantify the matrix similarity, thereby generating an efficient vector index and storing the index together with the original vector matrix in the vector database.

[0019] The step S3 specifically comprises:

[0020] S31 takes the task description as input and converts it into a vector matrix through vectorization technology. Then, ANN is used for retrieval, sorting the retrieved relevant documents, and selecting the most relevant documents or fragments as input to the language model.

[0021] S32 During the data generation phase, the relevant documents or fragments obtained will be used as input for language model training. Language models (such as GPT) are used to understand the grammatical structure, semantic relationships, and contextual information of the text. In this process, the model's self-attention mechanism can effectively capture long-distance dependencies within and between sentences, thereby deeply understanding the complex connections and potential patterns between information. Based on these understandings, the model can generate accurate natural language summaries and analysis texts according to specific task requirements.

[0022] The step S4 specifically includes: S41 takes the natural language summary and analysis text generated in the data generation phase as input, uses specific rules and templates to divide the input natural language text into blocks, and divides the blocks into structural modules such as "problems, solutions, decisions, and results".

[0023] S42 inputs the segmentation results and templates into the language model to automatically identify the structured elements in the text, automatically classify them into appropriate blocks, and automatically generate text summaries to extract keywords related to project success, challenges, risks, improvements, etc. for subsequent users.

[0024] The step S5 specifically comprises:

[0025] S51 creates a new workflow in Dify's workflow and defines its triggering conditions (such as user queries). Setting this workflow will automatically connect the retrieval and generation modules. When receiving a user query, the system first performs the retrieval and then outputs the answer through the generation model.

[0026] S52 configures the user interaction link in the Dify workflow to ensure that users can see the answers output from the retrieval and generation modules. At the same time, set up the necessary feedback mechanism to ensure that users can view the results or conduct subsequent interactions (such as modifying queries, asking new questions, etc.).

[0027] Wherein the step S6 specifically comprises:

[0028] S61 sets up scheduled task scheduling in Jenkins. According to actual needs, it sets the update frequency so that it can automatically start the data extraction process at the set time point, and then executes steps S1, S2, and S3 in sequence to generate a new summary of lessons learned.

[0029] A storage medium stores at least one instruction, which is loaded and executed by a processor to implement the method.

[0030] The advantages of the present invention are that the system loads data through the REST API interface of the relevant data source and converts the loaded data into a unified paradigm. Then, the system filters, cleans and repairs the loaded data and imports it into the system. After the data processing is completed, the system uses NLP technology to accurately segment the data and divide the data into sentences or words. Subsequently, the segmented data is vectorized to generate a corresponding vector matrix. Through the ANN algorithm, the system quantifies the matrix similarity, generates an efficient vector index, and stores the index together with the original vector matrix in the vector database. In the data analysis stage, the system uses RAG technology to retrieve information. First, the system takes the task description as input and calculates the similarity between the query vector and the document vector. Then, ANN is used for retrieval, sorts the retrieved related documents, and selects the most relevant documents or fragments as input and provides them to the language model. The language model understands the grammatical structure and semantic relationship of the text and generates a summary and analysis text of natural language. The generated natural language summary and analysis text is then organized and refined, and the text is divided into blocks using specific rules and templates. By training the model, the structured elements in the text are automatically identified, and a text summary is generated to extract keywords related to project success, challenges, risks, improvements, etc. The system creates a new workflow through the Dify chat tool and defines trigger conditions to automatically connect the retrieval and generation modules. When receiving a user query, the system first performs a search and then outputs the answer through the generation model. At the same time, the user interaction link is configured to ensure that users can view the results and conduct subsequent interactions. Finally, by setting the update frequency in Jenkins, the system can automatically start the data extraction process at the set time to ensure the real-time and accuracy of the lessons learned summary. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The following is a brief description of the contents expressed in the drawings of the present invention and the symbols in the drawings:

[0032] Figure 1 It is the overall architecture diagram of the present invention.

[0033] Figure 2 It is a flow chart of the present invention.

[0034] Figure 3 It is a model flow chart of step S1 of the present invention.

[0035] Figure 4 It is a process flow chart of step S1 of the present invention.

[0036] Figure 5It is a model flow chart of step S2 of the present invention.

[0037] Figure 6 It is a process flow chart of step S2 of the present invention.

[0038] Figure 7 It is a model flow chart of steps S3 and S4 of the present invention.

[0039] Figure 8 It is a process flow chart of steps S3 and S4 of the present invention.

[0040] Fig. 9 It is a process flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0041] The specific implementation of the present invention will be further explained in detail below by describing the optimal embodiment with reference to the accompanying drawings.

[0042] The present invention relates to the field of artificial intelligence and project management, and uses retrieval augmented generation (RAG) technology to realize project management experience and lessons analysis and automated integration method. The specific method includes: determining and extracting various data involved in the project management process from relevant data sources, and importing them into the system; segmenting the data imported into the system, then vectorizing the segmented data, and storing the data by building an efficient index mechanism; using RAG technology to analyze the data, in the retrieval stage, first using preset query conditions to retrieve information related to the current project from a stored database; in the generation stage, inputting the retrieved relevant data into a pre-trained language model, analyzing and understanding the information through the model, and generating Summary and analysis text in natural language; organize the generated summary text by using preset templates or structured methods to extract key lessons learned in the project management process; integrate the entire process into the automation tool, and by integrating the Dify chat tool, the generated experience can be obtained through chat and question-and-answer methods, thereby reducing the difficulty of experience acquisition; a regular update mechanism automatically extracts the latest data from project management tools such as Redmine to generate new summaries of lessons learned; the present invention uses intelligent technology to achieve efficient analysis and summary of project management lessons learned, and uses automation to improve the refinement and intelligence level of project management, thereby providing project managers with more accurate and real-time decision support.

[0043] The purpose of the present invention is to provide a project management experience and lessons analysis tool based on RAG technology and an automated integration method thereof. Combining the two functions of retrieval and generation of RAG technology, by first retrieving relevant information, then generating high-quality natural language text, and finally integrating it into the question-answering system, through intelligent chat, it can quickly provide project managers with accurate technical solutions, help them efficiently identify and solve potential problems in the project, avoid duplication, and realize the intelligent accumulation and application of experience and lessons, thereby significantly improving the decision-making quality and execution efficiency of project management. First, the original data is trained to obtain a summary and analysis text. Secondly, the lessons learned summary is integrated into the Dify chat tool. Finally, regular updates ensure the accuracy and timeliness of decisions.

[0044] The specific steps include:

[0045] S1 identifies and extracts various data involved in the project management process from relevant data sources and imports them into the system.

[0046] S2 segments the data imported into the system, then vectorizes the segmented data and stores the data by building an efficient indexing mechanism.

[0047] S3 uses RAG technology for data analysis. In the retrieval stage, the information related to the current project is first retrieved from the stored database using the preset query conditions. In the generation stage, the retrieved relevant data is input into the pre-trained language model, and the information is analyzed and understood by the model to generate a summary and analysis text in natural language.

[0048] S4 uses preset templates or structured methods to organize the generated summary text in a structured manner, extracting the key lessons learned in the project management process to facilitate user understanding.

[0049] S5 integrates the entire process into an automated tool. By integrating the Dify chat tool, the generated lessons learned summaries can be easily obtained through chat questions and answers, rather than relying on precise keyword searches. This reduces the difficulty of obtaining lessons learned while increasing the breadth of lessons learned.

[0050] S6 regularly updates its mechanism to automatically extract the latest data from relevant data sources and generate new lessons learned.

[0051] The present invention proposes a project management experience and lessons analysis tool based on RAG technology and its automated integration method, aiming to improve the efficiency and refinement level of project management through intelligent means. The system loads data through the RESTAPI interface of the relevant data source and converts the loaded data into a unified paradigm. Then, the system filters, cleans and repairs the loaded data and imports it into the system. After the data processing is completed, the system uses NLP technology to accurately segment the data and divide the data into sentences or words. Subsequently, the segmented data is vectorized to generate a corresponding vector matrix. Through the ANN algorithm, the system quantifies the matrix for similarity, generates an efficient vector index, and stores the index together with the original vector matrix in the vector database. In the data analysis stage, the system uses RAG technology for information retrieval. First, the system takes the task description as input and calculates the similarity between the query vector and the document vector. Then, ANN is used for retrieval, sorts the retrieved related documents, and selects the most relevant documents or fragments as input and provides them to the language model. The language model understands the grammatical structure and semantic relationship of the text and generates a summary and analysis text in natural language. The generated natural language summary and analysis text is then organized and refined, and the text is segmented using specific rules and templates. By training the model, the structured elements in the text are automatically identified, and a text summary is generated, extracting keywords related to project success, challenges, risks, improvements, etc. The system creates a new workflow through the Dify chat tool, and defines trigger conditions to automatically connect the retrieval and generation modules. When receiving a user query, the system first performs a search and then outputs the answer through the generation model. At the same time, the user interaction link is configured to ensure that users can view the results and conduct subsequent interactions. Finally, by setting the update frequency in Jenkins, the system can automatically start the data extraction process at the set time to ensure the real-time and accuracy of the lessons learned summary.

[0052] like Figure 1-8 As shown,

[0053] The present invention provides a project management experience and lessons analysis tool based on RAG technology and an automated integration method thereof, comprising:

[0054] S1 determines and extracts various data involved in the project management process from relevant data sources and imports them into the system;

[0055] The specific implementation is as follows:

[0056] S11 First, ensure access to the API interfaces of each data source, including applying for API keys, identity authentication and other steps to ensure data access under the premise of legal authorization. Secondly, use HTTP requests to obtain the required data from the data source, and specify the types of target data, such as project data, task data, and user data. Each time a request is made, it is necessary to ensure that the interface call meets expectations and that complete response data can be obtained. Finally, if the API returns a successful response, the returned JSON format data needs to be parsed, the relevant fields need to be extracted, and subsequent data processing or storage is performed to ensure that the data is in a unified format. The algorithm for accessing the data source is shown in Algorithm 1.

[0057]

[0058]

[0059] S12 After the data in S11 is acquired, the data needs to be deduplicated to ensure that the data items in each column do not contain duplicate items, and duplicate records are identified and deleted based on the unique identifier. The deduplicated data may also contain some abnormal data, so the data is further filtered and cleaned. First, the content in the project description column can be analyzed to identify and remove those fields that are not actually helpful for task analysis. Next, it is necessary to identify and remove abnormal values ​​in the data, such as data that does not comply with business rules such as dates set to future dates. For fields with missing values, they can be filled or repaired according to business rules or depending on the correlation of other fields to ensure the integrity and consistency of the data. Considering that the amount of data processed is large, it may lead to high memory usage. In order to improve system efficiency and reduce memory consumption, the data should be appropriately compressed, which helps save disk space, improve data reading speed, and optimize overall data processing performance. The data cleaning and optimization algorithm is shown in Algorithm 2.

[0060]

[0061] S2 segments the data imported into the system, then vectorizes the segmented data and stores the data by building an efficient index mechanism;

[0062] The specific implementation is as follows:

[0063] S21 uses the NLTK library to segment sentences and identify sentence boundaries based on punctuation marks (such as periods, question marks, and exclamation marks). The segmented sentences are segmented again into individual words based on space separation, punctuation marks, word form changes, etc. Finally, the words of all sentences are stored in a nested list.

[0064] S22 This method uses the ChatGPT-Embedding model as a vectorization tool, and generates an embedding vector that conforms to the ChatGPT style through the Embedding API provided by OpenAI. For each input sentence, first use the API to obtain its corresponding vector representation. These vectors accurately capture the semantic information of the sentence through high-dimensional space. At the same time, the K-means clustering algorithm is used to group the semantics of the sentence, classify sentences with similar semantics into one category, and improve the accuracy of semantic analysis. Subsequently, the vector representations of all sentences are aggregated into a matrix to facilitate further analysis of the model, and the generated vector is shown in Algorithm 3.

[0065]

[0066] S23 first takes the vector matrix as input and uses the FAISS (Facebook AI Similarity Search) algorithm to build an efficient approximate nearest neighbor search index. To further improve the search performance, the IndexIVF (Inverted File) technology is used to compress the index, thereby optimizing the query efficiency of large-scale data sets. By using the index built by FAISS, similarity queries can be quickly performed, returning the vectors most similar to the given query vector, and merging these similar vectors. The original vector matrix is ​​stored in the Pinecone database, so that in the subsequent retrieval process, the original vector most similar to the query vector can be quickly found. The specific algorithm is shown in Algorithm 4.

[0067]

[0068] S3 uses RAG technology for data analysis. In the retrieval stage, the information related to the current project is first retrieved from the stored database using the preset query conditions. In the generation stage, the retrieved relevant data is input into the pre-trained language model, and the information is analyzed and understood by the model to generate a summary and analysis text in natural language.

[0069] The specific implementation is as follows:

[0070] S31 takes the task description as input and converts it into a vector matrix through the ChatGPT-Embedding model. Next, the FAISS algorithm is used to build an index for the document vector, which can be quickly queried in the index, avoiding brute force traversal of all documents for similarity calculation, thereby speeding up the retrieval process. Using the retrieval function of the FAISS algorithm, the similarity indicators of each document are given: distance and index. Through these returned results, the documents can be sorted according to the similarity index, and the top k most similar documents can be selected as the final candidate input.

[0071] S32 In the data generation stage, the model will take the relevant documents or fragments obtained as input for training. Generate prompt words (Prompt) based on the relevant documents or fragments. In this method, the language model selects the GPT3.5 model as the training model for project experience summary. The GPT3.5 model itself does not have the ability to process sequence information, and position encoding is used to help the model understand the relative position of words in the sentence. Position encoding is usually generated by sine and cosine functions and added to word embedding to form the final input representation. Through the self-attention mechanism (Self-Attention), the model can interact with all other words in the sentence when processing each word to generate a weight for each word, thereby capturing the correlation between words. And the language model contains multiple encoder layers, each of which includes a multi-head self-attention mechanism and a feedforward neural network. The output of each layer will further adjust the representation of the word, so that the model can capture higher-level semantic information. Based on the output of these layers, the model will eventually classify or infer and generate the next most likely word until the text generation is completed.

[0072] S4 uses preset templates or structured methods to organize the generated summary text in a structured manner, extracting the key lessons learned in the project management process to facilitate user understanding.

[0073] The specific implementation is as follows:

[0074] S41 takes the natural language summary and analysis text generated in the data generation phase as input, and uses specific rules and templates to segment the input natural language text into blocks according to structural modules such as "problems, solutions, decisions, and results."

[0075] S42 inputs the segmentation results and templates into the language model to automatically identify the structured elements in the text, automatically classify them into appropriate blocks, and automatically generate text summaries to extract keywords related to project success, challenges, risks, improvements, etc. for subsequent users.

[0076] S5 integrates the entire process into an automated tool. By integrating the Dify chat tool, the generated lessons learned summaries can be easily obtained through chat questions and answers, rather than relying on precise keyword searches. This reduces the difficulty of obtaining lessons learned while increasing the breadth of lessons learned.

[0077] The specific implementation is as follows:

[0078] S51In the Dify workflow panel, create a workflow and set the trigger condition of the workflow to user query. When the system receives the text query entered by the user, the workflow is started. In the workflow design interface, add a "retrieval" node. Configure the retrieval module and the Pinecone data source to perform retrieval based on the query. In the workflow, create a "generate" node, select the trained GPT3.5 model, and pass the text returned by the retrieval module (that is, the relevant document fragments or knowledge content) as input to the generation model. Configure the output format of the generation model to ensure that the output is modularized according to structures such as "problems, solutions, decisions, and results."

[0079] S52In the workflow, add a "response" node, which is responsible for returning the generated answer to the user page. Add a "feedback" node to collect user feedback on the answer. The collected feedback can be used to adjust the behavior of subsequent workflows, such as improving the response quality of the retrieval algorithm or the generation model. After setting the feedback, the workflow should allow users to perform subsequent operations. Regularly collect user interaction data and feedback, analyze which problems occur frequently, and which links have problems (such as inaccurate retrieval, unclear answers from the generation model, etc.). Adjust the algorithm or workflow based on the feedback, and gradually optimize the entire system.

[0080] S6 regularly updates its mechanism to automatically extract the latest data from relevant data sources and generate new lessons learned.

[0081] The specific implementation is as follows:

[0082] S61 In the Jenkins Job configuration interface, in the "Build Trigger" section, select "Build periodically" and set a scheduled schedule (such as executing once every day at 12:00 midnight: H 0***). In the build section of the Jenkins task, configure the S1, S2, and S3 steps to ensure that each step obtains data, processes data, and trains the model in sequence. If any step fails, configure the Jenkins task to stop execution when it fails.

[0083] The present invention realizes efficient analysis and summary of project management experience and lessons through intelligent technology, and improves the refinement and intelligence level of project management through automated means, thereby providing project managers with more accurate and real-time decision support.

[0084] Obviously, the specific implementation of the present invention is not limited to the above-mentioned methods. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention, they are all within the protection scope of the present invention.

Claims

1. A project management experience processing method based on RAG, characterized by: The processing method comprises: S1. Extract various data involved in the project management process; S2: The data is segmented and then vectorized, and the data is stored through an index mechanism. S3. Use RAG technology to analyze data. In the retrieval stage, use the preset query conditions to retrieve information related to the current project from the stored database; in the generation stage, input the retrieved relevant data into the pre-trained language model, analyze and understand the information through the model, and generate a natural language summary and analysis text; S4. Use a preset template or structured approach to organize the generated summary text and extract the key lessons learned in the project management process; S5. Obtain user needs through the chatbot tool and obtain corresponding lessons learned data based on step 4 and present them to the user in a question-and-answer format.

2. A RAG-based project management experience processing method as claimed in claim 1, characterized in that: The method further comprises step S6: setting a regular update mechanism to automatically extract the latest data from relevant data sources regularly, generate new lessons learned and store them.

3. A RAG-based project management experience processing method as claimed in claim 1 or 2, characterized in that: The step S1 further comprises: S11. Load data through the REST API interface of the project-related data source, select the data format according to the requirements, and convert all data into a unified paradigm when loading data; S12. Filter, clean and repair the loaded data, remove redundant data, and fill or delete incomplete data; at the same time, compress the loaded data to improve data analysis speed and save memory space to obtain processed data.

4. A RAG-based project management experience processing method as claimed in claim 1 or 2, characterized in that: Step S2 specifically includes: S21. After processing the data, the natural language processing (NLP) technology is used to accurately segment the data, divide the text into sentences or words, and store the segmented fragments in a list in an orderly manner; S22, taking the segmented data as input, processing it using vectorization technology, converting it into a digital vector form, and generating a corresponding vector matrix; S23 takes the vector matrix as input and uses the approximate nearest neighbor ANN algorithm to quantify the matrix similarity, thereby generating an efficient vector index, and stores the index together with the original vector matrix in the vector database.

5. A RAG-based project management experience processing method as claimed in claim 1 or 2, characterized in that: Step S3 specifically includes: S31, taking the task description entered by the user through the chatbot tool as input, and converting it into a vector matrix through vectorization technology; then, using ANN to search, sort the retrieved relevant documents, and select the most relevant documents or fragments as input, and provide them to the pre-trained language model; S32. During the data generation phase, the relevant documents or fragments obtained will be used as input for language model training; the language model is used to understand the grammatical structure, semantic relations and contextual information of the text; the language model can generate accurate natural language summaries and analysis texts based on specific task requirements.

6. A RAG-based project management experience processing method as claimed in claim 1 or 2, characterized in that: The step S4 specifically includes: S41, taking the natural language summary and analysis text generated in the data generation phase as input, using preset rules and templates to divide the input natural language text into blocks according to structural modularization such as "problems, solutions, decisions, and results"; S42. Input the segmentation results and templates into the language model to automatically identify the structural elements in the text, automatically classify them into corresponding structural modular blocks, and automatically generate a text summary to extract keywords related to lessons learned for subsequent users.

7. A RAG-based project management experience processing method as claimed in claim 1 or 2, characterized in that: Step S5 specifically includes: the Dify chat tool is used as a chat robot tool, in the Dify workflow, a new workflow is created to define trigger conditions; the workflow is set to automatically connect the retrieval and generation modules; when a user query is received, the retrieval is performed first, and then the answer is output through the generation model.

8. A RAG-based project management experience processing method as claimed in claim 1 or 2, characterized in that: The step S6 specifically includes: setting a scheduled task schedule, setting an update frequency, so that it automatically starts the data extraction process at a set time point, and then executing steps S1, S2, and S3 in sequence to update and generate a new summary and analysis text.

9. A storage medium, characterized in that: The storage medium stores at least one instruction, and the instruction is loaded and executed by the processor to implement the method according to any one of claims 1-8.

Images