Domain technology tree generation method and device based on large model knowledge reasoning

Abstract

The invention belongs to the technical field of natural language processing and artificial intelligence, and discloses a domain technology tree generation method and device based on large model knowledge reasoning. The method comprises the following steps: retrieving the Internet and / or a field professional knowledge base by utilizing field technical keywords, obtaining and utilizing candidate retrieval contents and the field technical keywords to construct an instruction, and calling a field large model to judge the correlation between the candidate retrieval contents in the instruction and the technical keywords, obtaining and utilizing high-value candidate retrieval contents and field technology keywords to construct an instruction, and calling a field large model to extract a field technology triple from the instruction; and generating a domain technology tree by using the extracted domain technology triad. The key technology and the downstream industrial chain thereof are deeply mined by using the powerful domain knowledge background of the domain large model, and the downstream technology and the dependent technology of the key technology are mined at the same time, so that the completeness and the accuracy of key technology venation acquisition are effectively improved, and the difficulty of key technology information acquisition is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing and artificial intelligence technology, and in particular to a method and device for generating a domain technology tree based on large model knowledge reasoning. Background Art

[0002] In recent years, with the development of new energy, semiconductors, artificial intelligence and other high-tech technologies, it has become particularly important to sort out the key field nodes involved in the upstream, midstream and downstream of these technology industry chains to understand and grasp the laws and context of the development of the entire industry chain. Automated industry chain technology tree mining and generation need to rely on natural language processing methods to analyze massive scientific and technological text data to extract relevant technologies in the target field and the relationship between upstream and downstream technologies. With the development of machine learning and deep learning, methods such as intelligent optimization algorithms, deep neural network models and pre-trained language models have achieved leading performance in most natural language processing tasks such as text classification, named entity recognition, and relationship extraction. These methods provide good support for the mining of field technology trees. For example, for the analysis of the quantum chip (QuantumChip, QC) industry chain, the text in massive news and technical reports can be vectorized through the powerful representation ability of the pre-trained model, and relevant information can be quickly retrieved through semantic matching. And by performing entity recognition and relationship extraction on these text features, the upstream and downstream technology entities of QC are finally determined to complete the construction of the industry chain technology tree.

[0003] However, there are the following problems in constructing a technology tree using traditional machine learning and deep learning methods: First, the use of traditional machine learning and deep learning methods relies on expert wisdom, and the constructed technology tree is small in scale and shallow in depth, which makes it difficult to meet the research needs of deep decoupling analysis in the face of the ever-changing and evolving technology industry chain; second, scientific and technological information is iterating at a rapid pace. Faced with massive and complex information data with low value density, traditional technology tree construction methods find it difficult to mine high-value target technology information. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of existing technology tree construction, such as reliance on expert knowledge, small scale and shallow depth, and difficulty in mining high-value targets. A domain technology tree generation method based on large model knowledge reasoning is proposed to realize automated deep mining and correlation analysis of the hierarchical relationship of industrial chain product components in a specific technology field. The present invention also discloses a domain technology tree generation device based on large model knowledge reasoning, which provides users and scientific research institutions with a convenient and professional domain technology tree display method.

[0005] On the one hand, a method for generating a domain technology tree based on large model knowledge reasoning is provided, including:

[0006] Step 1: Use domain technology keywords to search the Internet and / or domain expertise database to obtain candidate search content;

[0007] Step 2: construct a first task instruction using the candidate search content and the technical keywords in the field, and call the domain big model to judge the relevance between the candidate search content and the technical keywords in the first task instruction to obtain high-value candidate search content;

[0008] Step 3: construct a second task instruction using the high-value candidate search content and the domain technology keywords, and call the domain big model to extract domain technology triples from the high-value candidate search content of the second task instruction, wherein the domain technology triples use the domain technology keywords as the subject, any word with "inclusion" semantics in the relationship set as the predicate, and the downstream technology keywords or dependent technology keywords as the object;

[0009] Step 4: Generate a domain technology tree using the extracted domain technology triples.

[0010] In an improved method for generating a domain technology tree based on large model knowledge reasoning, the method also includes:

[0011] Step 5: When it is determined that the extracted downstream technical keywords need to add child nodes, the extracted downstream technical keywords are used as domain technical keywords and jump to step 1.

[0012] In an improved method for generating a domain technology tree based on large model knowledge reasoning, step 1 includes:

[0013] Step 1-1: Use domain technology keywords to search the Internet and / or domain expertise database to obtain search content;

[0014] Step 1-2: Sort the search contents from high to low according to the relevance to the technical keywords in the field, and select the search contents with a set number of search contents ranked first as candidate search contents;

[0015] Step 1-3: Perform deduplication filtering and format analysis on the candidate search content to generate candidate search content in a unified format.

[0016] In an improved method for generating a domain technology tree based on large model knowledge reasoning, step 2 includes:

[0017] Step 2-1: traverse the candidate search contents in the candidate search data set, wherein all candidate search contents constitute the candidate search data set;

[0018] Step 2-2: The candidate search content and its corresponding field technology keywords are combined into key technology pairs, and the key technology pairs are combined into the first prompt question by using the prompt engineering;

[0019] Step 2-3: The first prompt question is used as the first task instruction and input into the domain model to obtain an answer as to whether the candidate search content matches the domain technical keywords;

[0020] Step 2-4: The candidate search contents answered as matching are regarded as high-value candidate search contents.

[0021] In an improved method for generating a domain technology tree based on large model knowledge reasoning, step 3 includes:

[0022] Step 3-1: construct a relationship set, wherein the relationship set includes predicate keywords representing the hierarchical relationship;

[0023] Step 3-2: traverse the high-value candidate search contents in the high-value candidate data set, wherein all high-value candidate search contents constitute the high-value candidate data set;

[0024] Step 3-3: Based on the high-value candidate search content, using the prompt engineering, the second prompt question is formed with the domain technology keyword corresponding to the high-value candidate search content as the subject and any word with "inclusion" semantics in the relationship set as the predicate;

[0025] Step 3-4: Use the second prompt question as the second task instruction and input it into the domain model to determine whether there is a predicate in the high-value candidate content that meets the triple technical relationship. When the judgment result is yes, generate a matching domain technology triple as an answer.

[0026] On the other hand, a domain technology tree generation device based on large model knowledge reasoning is provided, comprising:

[0027] A data retrieval module is used to search the Internet and / or a domain expertise database using domain technology keywords to obtain candidate search content;

[0028] A high-value content screening module is used to construct a first task instruction using the candidate search content and the field technical keywords, and call the field big model to judge the relevance between the candidate search content and the technical keywords in the first task instruction to obtain high-value candidate search content;

[0029] a technology triple extraction module, for constructing a second task instruction using the high-value candidate search content and the domain technology keywords, and calling the domain big model to extract domain technology triples from the high-value candidate search content of the second task instruction, wherein the domain technology triples have the domain technology keywords as the subject, any word with "inclusion" semantics in the relationship set as the predicate, and the downstream technology keywords or the dependent technology keywords as the object; and

[0030] The domain technology tree generation module is used to generate a domain technology tree using the extracted domain technology triples.

[0031] In an improved domain technology tree generation device based on large model knowledge reasoning, it also includes:

[0032] The domain technology tree display module is used to use the extracted downstream technology keywords as domain technology keywords to trigger the data retrieval module when it is determined that the extracted downstream technology keywords need to add sub-nodes.

[0033] In an improved domain technology tree generation device based on large model knowledge reasoning, the data retrieval module includes: a retrieval unit, a sorting unit and a filtering unit;

[0034] The retrieval unit is used to search the Internet and / or a domain expertise database using domain technology keywords to obtain retrieval content;

[0035] The ranking unit is used to rank the search contents from high to low according to the relevance to the technical keywords of the field, and take the search contents ranked first by a set number as candidate search contents;

[0036] The filtering unit is used to perform deduplication filtering operations and format analysis on the candidate search content to generate candidate search content in a unified format.

[0037] In an improved domain technology tree generation device based on large model knowledge reasoning, the high-value content screening module includes: a traversal unit, a first prompt question composition unit, a matching unit and a high-value content construction unit;

[0038] The traversal unit is used to traverse the candidate search contents in the candidate search data set, wherein all the candidate search contents constitute the candidate search data set;

[0039] The first prompt question forming unit is used to form a key technology pair with the candidate search content and its corresponding field technology keyword, and use the prompt engineering to form the key technology pair into the first prompt question;

[0040] The matching unit is used to input the first prompt question as the first task instruction into the domain macro model to obtain an answer as to whether the candidate search content matches the domain technical keyword;

[0041] The high-value content construction unit is used to take the candidate search content answered as matching as the high-value candidate search content.

[0042] In an improved domain technology tree generation device based on large model knowledge reasoning, the technology triple extraction module includes: a relationship set construction unit, a traversal unit, a second prompt question composition unit and a technology triple generation unit;

[0043] The relationship set construction unit is used to construct a relationship set, wherein the relationship set contains predicate keywords representing the hierarchical relationship;

[0044] The traversal unit is used to traverse the high-value candidate search contents in the high-value candidate data set, wherein all high-value candidate search contents constitute the high-value candidate data set;

[0045] The second prompt question composition unit is used to form a second prompt question based on the high-value candidate search content and using the prompt project, with the field technology keyword corresponding to the high-value candidate search content as the subject and any word with "inclusion" semantics in the relationship set as the predicate;

[0046] The technical triplet generation unit is used to use the second prompt question as the second task instruction, input it into the domain big model to determine whether there is a predicate that meets the triplet technical relationship in the high-value candidate content, and when the judgment result is that it exists, generate a matching domain technology triplet as an answer.

[0047] Compared with the prior art, the advantages of the present invention are:

[0048] 1. The embodiment of the present invention provides a method for obtaining high-quality search information, which realizes the retrieval of technical texts in a simple and efficient manner by combining knowledge base retrieval, keyword matching, semantic similarity calculation, relevance ranking and other search methods. At the same time, by combining the cross-domain knowledge background of the domain big model, the search content is further judged and screened, ensuring the acquisition of high-value search content that matches the keywords;

[0049] 2. The embodiment of the present invention aims at the problem that it is difficult to decouple the key and difficult points of the field technology in the traditional technology tree generation. By adopting the extraction of technology triples as the main basis for technology tree generation, a more robust technology tree generation method is provided. At the same time, by using the powerful domain knowledge background of the domain big model, the key technology and its downstream industrial chain are deeply mined, and the downstream technology and dependent technology of the key technology are mined at the same time, which effectively improves the completeness and accuracy of the key technology context acquisition and greatly reduces the difficulty of obtaining key technology information;

[0050] 3. The solution of the embodiment of the present invention utilizes a domain big model with strong domain background knowledge on the one hand, and first extracts high-value candidate retrieval content from the candidate retrieval content on the other hand, and then extracts technical triples from the high-value candidate retrieval content, so that non-high-value candidate retrieval content that accounts for a large proportion of the overall data volume is eliminated from the beginning. Compared with the solution of using BERT and other models for small tasks, the solution of first extracting technical triples from candidate retrieval content and then performing normalization operations such as denoising on the technical triples greatly reduces the amount of data processing for triple extraction, and because the domain big model relies on large parameters and rich background knowledge, the screening of high-value candidate retrieval content is more accurate, and the quality of retrieval content is higher, which helps to improve the extraction quality of technical triples from the source of the data, thereby improving the quality of the generated domain data tree. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is one of the flow charts of the method for generating a domain technology tree based on large model knowledge reasoning according to an embodiment of the present invention;

[0052] Figure 2 This is the second flow chart of the method for generating a domain technology tree based on large model knowledge reasoning according to an embodiment of the present invention;

[0053] Figure 3 It is a key technology of the embodiment of the present invention to retrieve data samples;

[0054] Figure 4 It is a sample of high-value retrieval content and technology triples extracted in an embodiment of the present invention;

[0055] Figure 5 It is a schematic diagram of the structure of a domain technology tree generation device based on large model knowledge reasoning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In the scheme of the embodiment of the present invention, the first prompt question and the second prompt question are constructed respectively by using domain technology keywords and two search contents. Correspondingly, the same domain big model with strong domain background knowledge is called twice to generate high-value candidate search contents and extract domain technology triples respectively, so that it is possible to mine high-quality domain technology trees from the search contents to meet the growing demand for domain technology tree construction.

[0057] The domain big model in the embodiment of the present invention is a model obtained by fine-tuning the trained big model with instructions using a professional knowledge base in a specific field. For example, the domain big model is obtained by fine-tuning the already trained open source big models such as Tongyi Qianwen, Baichuan, or ChatGLM3 using a corpus of knowledge including various aspects such as the quantum field. At this time, the solution of the embodiment of the present invention can use the domain big model to construct a domain technology tree for cutting-edge technologies including the quantum field technology tree.

[0058] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0059] like Figure 1 As shown, it is a flow chart of a method for generating a domain technology tree based on large model knowledge reasoning provided by an embodiment of the present invention, comprising the following steps:

[0060] Step 1: Obtain the domain technology keywords needed to construct the domain technology tree, use the domain technology keywords to search the Internet and / or domain expertise database, perform data preprocessing operations such as format constraints on the search content, and obtain candidate search content;

[0061] Specifically, this step 1 may include steps 1-1 to 1-3:

[0062] Step 1-1 Use domain technology keywords i , search the Internet and / or domain expertise database to obtain the search content v i ;

[0063] The search content obtained here can be paragraphs or text fragments.

[0064] The retrieval process is to identify and obtain information system resources related to the information need. Each knowledge keyword k i Corresponding to the knowledge text v i .

[0065] Taking quantum chips as an example, by searching for "components of quantum chips" and "dependencies of quantum chips", we obtain several text fragments containing quantum chips. i , such as "a cooperation plan launched by the European Union and its member states at the end of 2018... Among them, the core projects in the field of quantum technology and human brain mapping include the research and development and application of quantum chips..."

[0066] Step 1-2: Retrieve content v i Re-sort according to the technical keywords i The relevance of the candidate search content v′ is obtained from high to low i .

[0067] Given the technical keyword q that needs to be queried i , use the similarity function s to search for the most similar top k keys and obtain the text content v′ of the paired value i .

[0068] Step 1-3: Search candidate content v′ i Perform deduplication filtering and format parsing to generate candidate search content v′ in a unified format i .

[0069] Continuing with the above example, remove the search paragraphs that do not include "quantum chip" and remove the candidate search paragraphs that contain illegal characters.

[0070] In this step 1-3, text information and repeated content that do not contain technical keywords in the search content are eliminated, and the candidate data can be converted into a list format that is easy to process, and finally the candidate search data set D is obtained. cand .like Figure 3 In the example shown in , the technical keyword to be queried is "Extreme Ultraviolet Lithography Machine". After searching the knowledge base and screening and sorting, the final candidate search data set is ["Capillary discharge extreme ultraviolet light source extreme ultraviolet lithography is considered to be the preferred technology for the next generation of lithography, and will replace optical lithography after 2015 to become the mainstream lithography technology in semiconductor production. The extreme ultraviolet lithography system is a comprehensive high-tech...", "Ten. Ultraviolet lithography technology has made a breakthrough. In February 2020, the Belgian Microelectronics Research Center and ASML of the Netherlands used the NXE:3400B extreme ultraviolet lithography equipment..."].

[0071] Step 2: Use the candidate search content and the field technical keywords to construct a first task instruction, call the domain big model to judge the relevance between the candidate search content and the technical keywords in the first task instruction, and obtain high-value candidate search content.

[0072] In this step 2, the retrieved content and keywords are combined to form the first task instruction, and the domain big model is used to judge the relevance of the instruction, filtering and retaining high-value search information. The first task instruction here is used as the big model input judgment instruction; then the domain big model is used to judge each task instruction to determine whether the current search content is related to the technical keywords of the given query. If it is related, it will be used as a high-value candidate search content.

[0073] Specifically, this step 2 may include steps 2-1 to 2-4:

[0074] Step 2-1: Traverse the dataset D consisting of all candidate search contents cand Candidate search content v′ in i ;

[0075] Step 2-2: Set v′i The corresponding technical keywords i Key technology components <v′ i ,q i >, use the prompt project to form the first prompt question S i ;

[0076] like Figure 4 As shown, the first prompt question constructed by the technical keyword "extreme ultraviolet lithography machine" and the search content "X. Ultraviolet lithography technology has achieved breakthroughs. In February 2020, the Belgian Microelectronics Research Center and ASML of the Netherlands used NXE:3400B extreme ultraviolet lithography equipment..." can be "Assuming that you are a very good industry chain analyst, please determine whether the following candidate sentences contain content related to the downstream technology of quantum chips. If so, answer True, if not, please answer False, and give an explanation."

[0077] Step 2-3: Set the first prompt question S i As the first task instruction, call the domain model to use the following formula to process and obtain the candidate search content v′ i and technical keywords q i whether the answer matches;

[0078] Specifically, the sampling decoding method can be used to call the large model to generate the corresponding decision answer. According to the autoregressive decoding method of the large model, the decision answer is obtained according to the following formula.

[0079]

[0080] Among them, P is the probability of generating the next word, S i is the character vector of all the text in the prompt question, T is the length limit of the domain model generation, w is the character currently generated during the model generation process, I is the sequence number of the currently generated word, and t is a random variable ranging from 1 to the longest number of words generated in the sentence. Step 2-4: The answer is the matching candidate search content Retrieve content as high-value candidates.

[0081] Here, the dataset consisting of all high-value candidate retrieval contents can be used as the high-value candidate dataset D hv .

[0082] Due to the powerful knowledge reserve and natural language reasoning ability of the large model, the large model is used to determine whether the current search information is related to the key technology, so as to screen out high-value candidate data sets that are closely related to the key technology. hv .

[0083] Continuing with the above example, "In the above text, the main components of the "quantum chip" are mentioned, so True is returned. The explanation is as follows: A quantum computer is a computer based on the principles of quantum mechanics. Its basic unit is the quantum bit (qubit). Unlike the bits of traditional computers, quantum bits can be in multiple states at the same time, which makes quantum computers more efficient and have stronger computing power than traditional computers when dealing with certain specific problems."

[0084] Step 3: Use the high-value candidate search content and the domain technology keywords to construct a second task instruction, and call the domain big model to extract domain technology triples from the high-value candidate search content of the second task instruction.

[0085] Among them, the domain technology triplet takes the domain technology keyword as the subject, any word with the "inclusion" semantics in the relationship set as the predicate, and the downstream technology keyword or the dependent technology keyword as the object.

[0086] In this step 3, for the filtered high-value information, instructions are constructed in combination with prompts such as technical keywords, and the domain big model is called to extract key technology triples, that is, hierarchical node mining and dependency node mining of key technologies. The extracted results can be preprocessed to filter out unreasonable triples.

[0087] Specifically, this step 3 may include steps 3-1 to 3-4:

[0088] Step 3-1: construct a relationship set, wherein the relationship set includes predicate keywords representing the hierarchical relationship;

[0089] Step 3-2: Traverse the high-value candidate dataset D hv High-value candidate search content in;

[0090] Step 3-3: Based on the high-value candidate search content, use the prompt engineering to form the second prompt question with the domain technology keyword corresponding to the high-value candidate search content as the subject and any word with "inclusion" semantics in the relationship set as the predicate

[0091] Using the example of the two-word chip above, the second prompt question generated here is It can be: "Assuming that you are a very good industry chain analyst, please extract the downstream technology of quantum chips in the given text and return it in the form of triples in JSON format. The triple format is {"head":quantum chip,"relation":constituent components,"tail":downstream technology}".

[0092] Step 3-4: Add the second prompt question As the second task instruction, it calls the domain model to determine high-value candidate content Whether there is a predicate that meets the triple technical relationship, when the judgment result is yes, generate a matching technical triple as an answer.

[0093] Continuing with the example of quantum chip, the dependent technology of quantum chip successfully extracted here is the dependent technology triple of control electronic device, which is "{'head':'quantum chip','relation':'depends on','tail':'control electronic device'}".

[0094] like Figure 4 In the example shown, given the technical keyword "Extreme Ultraviolet Lithography Machine" and one of its corresponding high-quality search information contents "X. Ultraviolet lithography technology has achieved a breakthrough. In February 2020, the Belgian Microelectronics Research Center and ASML of the Netherlands used NXE:3400B extreme ultraviolet lithography equipment...", The content is "Extract the relationship triplets about the components of the extreme ultraviolet lithography machine in the following text, and answer \n{input text} in the format of JSON string [{'head':",'relation':",'tail':"},…]". After being processed by the large model, an answer is generated, the content is "In the above text, the main components of the extreme ultraviolet lithography machine are mentioned, so True is returned. Specifically, the following key components are mentioned in the text: 1. Extreme Ultraviolet Co., Ltd. (EUVLLC)…”. At the same time, the natural language answers generated by the large model are formatted and parsed, and the answers that do not conform to the JSON format and the given subject and predicate are eliminated. The final technical triples are: [{'head':'Extreme Ultraviolet Lithography Machine','relation':'Components','tail':'EUV Light Source'},{'head':'Extreme Ultraviolet Lithography Machine','relation':'Components','tail':'Illumination Optical System'},{'head':'Extreme Ultraviolet Lithography Machine','relation':'Components','tail':'Microprojection Optical System'}…].

[0095] Step 4: Construct a technology tree using the extracted technology triples.

[0096] Step 5: When it is determined that the extracted downstream technical keywords need to add child nodes, the extracted downstream technical keywords are used as domain technical keywords and jump to step 1.

[0097] Specifically, by inputting key technologies, such as quantum chips, the method will complete steps 1 to 4, execute step 5 for iteration, and display a complete field technology tree about quantum chips in the form of a multi-branch tree.

[0098] At the same time, the nodes in the technology tree include downstream technology nodes and dependent technology nodes, which can clearly show the development context of technology. For example, the downstream technology nodes of quantum chips include: photonic chip methods, quantum dots, semiconductor qubits, metamaterial antennas and quantum coherent receivers, etc. The dependent technologies of quantum chips include: quantum bits, superconducting quantum, silicon electronics, silicon nitride photonics, etc. In addition, all nodes generated by this method can be added, deleted and displayed in detail, such as Figure 2 As shown, the details include the source of the search content corresponding to the node and the explanation of the domain model.

[0099] The embodiment of the present invention provides a method for generating a domain technology tree based on big model knowledge reasoning. By using this method to mine and generate a technology tree for key technologies, the understanding of the context of key technologies in the domain can be effectively improved. Specifically, the method of the present invention is used to retrieve text related to key technologies, and the domain big model is used to perform high-value content judgment on candidate retrieval content. In addition, the domain big model is used to generate downstream technology and dependent technology triples that are closely related to the content, thereby accurately constructing a complete domain technology tree related to key technologies.

[0100] In addition, the embodiment of the present invention also provides a domain technology tree generation device based on large model knowledge reasoning, such as Figure 5 As shown, it includes: a data retrieval module 51, a high-value content screening module 52, a technology triple extraction module 53 and a domain technology tree generation module 54; wherein:

[0101] The data retrieval module 51 is used to search the Internet and / or the domain expertise database using domain technology keywords to obtain candidate search content;

[0102] A high-value content screening module 52 is used to construct a first task instruction using the candidate search content and the field technical keywords, and call the field big model to judge the relevance between the candidate search content and the technical keywords in the first task instruction to obtain high-value candidate search content;

[0103] A technology triple extraction module 53 is used to construct a second task instruction using the high-value candidate search content and the domain technology keywords, and call the domain big model to extract a domain technology triple from the high-value candidate search content of the second task instruction, wherein the domain technology triple has the domain technology keyword as the subject, any word with "inclusion" semantics in the relationship set as the predicate, and the downstream technology keyword or the dependent technology keyword as the object; and

[0104] The domain technology tree generation module 54 is used to generate a domain technology tree using the extracted domain technology triples.

[0105] Preferably, it also includes:

[0106] The domain technology tree display module 55 is used to use the extracted downstream technology keywords as domain technology keywords to trigger the data retrieval module when it is determined that the extracted downstream technology keywords need to add child nodes.

[0107] Specifically, the data retrieval module 51 includes: a retrieval unit, a sorting unit and a filtering unit;

[0108] The retrieval unit is used to search the Internet and / or a domain expertise database using domain technology keywords to obtain retrieval content;

[0109] The ranking unit is used to rank the search contents from high to low according to the relevance to the technical keywords of the field, and take the search contents ranked first by a set number as candidate search contents;

[0110] The filtering unit is used to perform deduplication filtering operations and format analysis on the candidate search content to generate candidate search content in a unified format.

[0111] Preferably, the high-value content screening module 52 includes: a traversal unit, a first prompt question composition unit, a matching unit and a high-value content construction unit;

[0112] A traversal unit, used to traverse the candidate search contents in the candidate search data set, wherein all the candidate search contents constitute the candidate search data set;

[0113] The first prompt question forming unit is used to form a key technology pair with the candidate search content and its corresponding field technology keyword, and form the key technology pair into a first prompt question by using the prompt engineering;

[0114] A matching unit, used to input the first prompt question as a first task instruction into the domain macro model, and obtain an answer as to whether the candidate search content matches the domain technical keyword;

[0115] The high-value content construction unit is used to take the candidate retrieval content answered as matching as the high-value candidate retrieval content.

[0116] Specifically, the technical triple extraction module 53 includes: a relationship set construction unit, a traversal unit, a second prompt question composition unit and a technical triple generation unit;

[0117] The relationship set construction unit is used to construct a relationship set, wherein the relationship set contains predicate keywords representing the hierarchical relationship;

[0118] The traversal unit is used to traverse the high-value candidate search content in the high-value candidate data set;

[0119] The second prompt question composition unit is used to form a second prompt question based on the high-value candidate search content and using the prompt project, with the field technology keyword corresponding to the high-value candidate search content as the subject and any word with "inclusion" semantics in the relationship set as the predicate;

[0120] The technical triplet generation unit is used to use the second prompt question as the second task instruction, input it into the domain big model to determine whether there is a predicate that meets the triplet technical relationship in the high-value candidate content, and when the judgment result is that it exists, generate a matching domain technology triplet as an answer.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention and should be included in the scope of the claims of the present invention.

Claims

1. A method for generating a domain technology tree based on large model knowledge reasoning, comprising: Step 1: Use domain technology keywords to search the Internet and / or domain expertise database to obtain candidate search content; Step 2: construct a first task instruction using the candidate search content and the technical keywords in the field, and call the domain big model to judge the relevance between the candidate search content and the technical keywords in the first task instruction to obtain high-value candidate search content; Step 3: construct a second task instruction using the high-value candidate search content and the domain technology keywords, and call the domain big model to extract domain technology triples from the high-value candidate search content of the second task instruction, wherein the domain technology triples use the domain technology keywords as the subject, any word with "inclusion" semantics in the relationship set as the predicate, and the downstream technology keywords or dependent technology keywords as the object; Step 4: Generate a domain technology tree using the extracted domain technology triples.

2. The method for generating a domain technology tree based on large model knowledge reasoning according to claim 1 is characterized in that: Also includes: Step 5: When it is determined that the extracted downstream technical keywords need to add child nodes, the extracted downstream technical keywords are used as domain technical keywords and jump to step 1.

3. The method for generating a domain technology tree based on large model knowledge reasoning according to claim 1 is characterized in that: Step 1 includes: Step 1-1: Use domain technology keywords to search the Internet and / or domain expertise database to obtain search content; Step 1-2: Sort the search contents from high to low according to the relevance to the technical keywords in the field, and select the search contents with a set number of search contents ranked first as candidate search contents; Step 1-3: Perform deduplication filtering and format analysis on the candidate search content to generate candidate search content in a unified format.

4. The method for generating a domain technology tree based on large model knowledge reasoning according to any one of claims 1 to 3, characterized in that: Step 2 includes: Step 2-1: traverse the candidate search contents in the candidate search data set, wherein all candidate search contents constitute the candidate search data set; Step 2-2: The candidate search content and its corresponding field technology keywords are combined into key technology pairs, and the key technology pairs are combined into the first prompt question by using the prompt engineering; Step 2-3: The first prompt question is used as the first task instruction and input into the domain model to obtain an answer as to whether the candidate search content matches the domain technical keywords; Step 2-4: The candidate search contents answered as matching are regarded as high-value candidate search contents.

5. The method for generating a domain technology tree based on large model knowledge reasoning according to claim 4 is characterized in that: Step 3 includes: Step 3-1: construct a relationship set, wherein the relationship set includes predicate keywords representing the hierarchical relationship; Step 3-2: traverse the high-value candidate search contents in the high-value candidate data set, wherein all high-value candidate search contents constitute the high-value candidate data set; Step 3-3: Based on the high-value candidate search content, using the prompt engineering, the second prompt question is formed with the domain technology keyword corresponding to the high-value candidate search content as the subject and any word with "include" semantics in the relationship set as the predicate; Step 3-4: Use the second prompt question as the second task instruction and input it into the domain model to determine whether there is a predicate in the high-value candidate content that meets the triple technical relationship. When the judgment result is yes, generate a matching domain technology triple as an answer.

6. A domain technology tree generation device based on large model knowledge reasoning, characterized in that: include: A data retrieval module is used to search the Internet and / or a domain expertise database using domain technology keywords to obtain candidate search content; A high-value content screening module is used to construct a first task instruction using the candidate search content and the field technical keywords, and call the field big model to judge the relevance between the candidate search content and the technical keywords in the first task instruction to obtain high-value candidate search content; A technology triple extraction module is used to construct a second task instruction using the high-value candidate search content and the field technology keywords, and call the domain big model to extract a field technology triple from the high-value candidate search content of the second task instruction, wherein the field technology triple takes the field technology keyword as the subject, any word with "inclusion" semantics in the relationship set as the predicate, and the downstream technology keyword or the dependent technology keyword as the object; and The domain technology tree generation module is used to generate a domain technology tree using the extracted domain technology triples.

7. The domain technology tree generation device based on large model knowledge reasoning according to claim 6 is characterized in that: Also includes: The domain technology tree display module is used to use the extracted downstream technology keywords as domain technology keywords to trigger the data retrieval module when it is determined that the extracted downstream technology keywords need to add sub-nodes.

8. The domain technology tree generation device based on large model knowledge reasoning according to claim 6 is characterized in that: The data retrieval module includes: a retrieval unit, a sorting unit and a filtering unit; The retrieval unit is used to search the Internet and / or a domain expertise database using domain technology keywords to obtain retrieval content; The ranking unit is used to rank the search contents from high to low according to the relevance to the technical keywords of the field, and take the search contents ranked first by a set number as candidate search contents; The filtering unit is used to perform deduplication filtering operations and format analysis on the candidate search content to generate candidate search content in a unified format.

9. The domain technology tree generation device based on large model knowledge reasoning according to any one of claims 6 to 8 is characterized in that: The high-value content screening module includes: a traversal unit, a first prompt question composition unit, a matching unit and a high-value content construction unit; The traversal unit is used to traverse the candidate search contents in the candidate search data set, wherein all the candidate search contents constitute the candidate search data set; The first prompt question forming unit is used to form a key technology pair with the candidate search content and its corresponding field technology keyword, and use the prompt engineering to form the key technology pair into the first prompt question; The matching unit is used to input the first prompt question as the first task instruction into the domain macro model to obtain an answer as to whether the candidate search content matches the domain technical keyword; The high-value content construction unit is used to take the candidate search content answered as matching as the high-value candidate search content.

10. The domain technology tree generation device based on large model knowledge reasoning according to claim 9 is characterized in that: The technical triple extraction module includes: a relationship set construction unit, a traversal unit, a second prompt question composition unit and a technical triple generation unit; The relationship set construction unit is used to construct a relationship set, wherein the relationship set contains predicate keywords representing the hierarchical relationship; The traversal unit is used to traverse the high-value candidate search contents in the high-value candidate data set, wherein all high-value candidate search contents constitute the high-value candidate data set; The second prompt question composition unit is used to form a second prompt question based on the high-value candidate search content and using the prompt project, with the field technology keyword corresponding to the high-value candidate search content as the subject and any word with "include" semantics in the relationship set as the predicate; The technical triplet generation unit is used to use the second prompt question as the second task instruction, input it into the domain big model to determine whether there is a predicate that meets the triplet technical relationship in the high-value candidate content, and when the judgment result is that it exists, generate a matching domain technology triplet as an answer.

Images