Innovative design inspiration stimulation method based on patent semantic network concept combination generation

Abstract

The present invention relates to an innovative design inspiration motivation method based on patent semantic network concept combination generation, comprising the following steps: limiting a patent set used to construct a patent semantic memory network according to target functional verbs of design requirements, constructing a main function expression template for patent texts according to the expression form of the main functions in the patents; extracting main functional verbs from selected patent data based on the main functional expression templates; acquiring a target functional patent text set based on a semantic dictionary: after determining the main functional verbs of each patent, acquiring target functional patents used to construct a semantic memory network according to the semantic similarity between the main functional verbs of the patents and the target functional verbs, and forming a target functional patent text set; constructing a patent semantic memory network based on semantic computing; and inspiring inspiration of long-distance concept combinations.

Description

Technical Field

[0001] The present invention belongs to the field of information modeling based on semantics and patent knowledge, and relates to a method for constructing a patent semantic knowledge network, and specifically to an innovative design inspiration motivation method based on the generation of patent semantic network concept combinations. Background Art

[0002] Conceptual design is a process of solving design problems through the interaction of innovative design thinking and design knowledge. In this process, designers need to give full play to their creativity and comprehensively apply their own design knowledge and experience to find possible solutions. At the same time, designers need to break through the limitations of their own design knowledge and experience and expand the knowledge space for searching solutions[1]. This process not only places high demands on the designer's innovative ability, but also requires the designer to have sufficient knowledge reserves. As a product of innovation, patent texts contain the latest and most innovative knowledge and are an important knowledge resource for designers to expand their design knowledge space[2]. However, manual retrieval and reading of patents by designers brings a great cognitive burden to designers and requires a lot of time and energy. Patent analysis-related technologies have emerged. Through information technologies such as artificial intelligence, big data, and machine learning, patents can be efficiently retrieved and knowledge extracted from them [3]. Liu et al. proposed a function-based patent knowledge retrieval tool that generates patent knowledge space by extracting patent function information, technical terms, and classification information, and divides patent fields by clustering to achieve cross-domain patent recommendations, thereby stimulating designers' creativity and supporting product innovation concept design [4]. Wang et al. proposed a method for extracting principle innovation knowledge from process patents and a framework for principle knowledge extraction. They rationally organized principle innovation knowledge by establishing a formal representation model of the process contradiction matrix [5]. Kim proposed a semi-automatic method for extracting technical causal relationships from patents to assist new product development and technical opportunity discovery [6]. Luo et al. constructed an overall technology space map based on the technical fields of a large number of patents, covering the entire technology space, allowing designers to conceive new design opportunities beyond their knowledge scope [7]. Although patent analysis technology reduces the burden of designers' manual knowledge retrieval, designers' use of patent and design knowledge is still limited by their own innovation and cognitive abilities. In the field of patent knowledge-assisted design, existing research focuses on related patent recommendation and patent knowledge extraction. Although the former saves designers the time of searching patents, designers still need to spend time to extract design knowledge from patents; although the latter helps designers extract design knowledge, the design knowledge in existing research is mainly expressed in the form of keywords, which is difficult to fully express the design knowledge in patents, and its auxiliary effect on design is very limited. Classical cognitive theory proposes that semantic memory can be represented in the form of a semantic memory network, where nodes represent various concepts and edges represent the semantic relationship between different concepts. The semantic relationship between different concepts is usually connected by semantic similarity of different strengths, and different strengths reflect the semantic distance between concepts [8]. The massive patent documents contain rich design knowledge. Using the massive patents and the design knowledge contained therein to establish a patent semantic memory network can break through the limitations of individual semantic memory of designers.In addition, in actual product concept design, designers usually use products with the same or similar functions as design references. Therefore, obtaining functional knowledge from massive patents can better recommend innovative concepts and support designers in forming creative concepts or plans, which is of great significance to improving the company's product innovation capabilities and product competitiveness.

[0003] This patent aims to support concept generation based on functional knowledge extraction, and proposes a patent semantic memory network model based on the semantic memory network; and uses natural language processing (NLP) technology and network science methods to establish a patent semantic memory network, providing designers with concept combination incentives to assist them in generating design concepts, helping designers to comprehensively use internal and external knowledge to generate innovative concepts and solutions, and promote product innovation design.

[0004] References

[0005] [1]SARICA S,LUO J,WOOD K L.TechNet:Technology semantic network based patent data[J].Expert Systems with Applications,2020,142:112995.

[0006] [2] Cao Shujin, Li Ruijing. Construction and application of innovation knowledge graph based on patent document abstracts[J]. Information Theory and Practice, 2022, 45(11): 21-28.

[0007] [3] Liu Chunjiang, Zhu Jiang. Research on the architecture of patent big data service platform for intelligence analysis[J]. Library Work and Research, 2022(04):57-64.

[0008] [4]Liu L, Li YA, Xiong YA, et al. A new function-based patent knowledgeretrieval tool for conceptual design of innovative products[J]. Computers in Industry, 2020,115:103154.

[0009] [5]Wang GF, Tian XT, Geng JH, et al. Extraction of principle knowledge from process patents for manufacturing process innovation[J]. Procedia CIRP, 2016,56:193-198.

[0010] [6]Kim H,Joung J,Kim K.Semi-automatic extraction of technologicalcausality from patents[J].Computers&Industrial Engineering,2018,115:532-542.

[0011] [7] Luo J, Song B, Blessing L, et al. Design opportunity conception using the total technology space map [J]. Artificial Intelligence for Engineering Design Analysis and Manufacturing, 2018, 32(4): 449-461.

[0012] [8]Collins AM, Loftus E FA spreading-activation theory of semantic processing[J]. Readings in Cognitive Science, 1988, 82(6):126-136. Summary of the invention

[0013] In order to assist designers in concept stimulation and solve the problems existing in existing patent knowledge mining technology, the present invention aims to provide an innovative design inspiration stimulation method based on patent semantic network concept combination generation. The main technical solutions are as follows:

[0014] A method for stimulating innovative design inspiration based on patent semantic network concept combination generation, comprising the following steps:

[0015] Step 1: According to the target functional verbs of the design requirements, the patent set used to construct the patent semantic memory network is limited, and the main function expression template for the patent text is constructed according to the expression form of the main functions in the patent;

[0016] Step 2: Extract the main functional verbs from the selected patent data based on the main functional expression template;

[0017] Step 3: Acquisition of target functional patent text set based on semantic dictionary: After determining the main functional verbs of each patent, the target functional patents for constructing the semantic memory network are acquired according to the semantic similarity between the main functional verbs of the patent and the target functional verbs, and the target functional patent text set is formed;

[0018] Step 4: Constructing a patent semantic memory network based on semantic computing, the method is as follows:

[0019] ⑴ Patent concept extraction based on TF-IDF and part of speech. Patent concept refers to the keyword that can express the content of the patent subject. Therefore, the patent texts in the patent target function text set are firstly subjected to keyword concept extraction, and then summarized into a target function patent concept set composed of keyword concepts. The keyword extraction is performed using the TF-IDF algorithm.

[0020] (2) Calculate the comprehensive semantic similarity based on the method based on semantic dictionary and corpus: After extracting all the concept words in the target functional patent text set and determining the nodes in the semantic memory network, the nodes are connected by edges to form a complete semantic memory network. The edges in the semantic memory network are the semantic similarities between the two concept words corresponding to the nodes; the comprehensive semantic similarity is calculated by the method based on semantic dictionary and corpus;

[0021] (3) Connect each concept pairwise to construct a fully connected concept semantic memory network;

[0022] Step 5: Inspiration and motivation of long-distance concept combination. The method is as follows: according to the design task requirements, concepts are selected from the fully connected concept semantic memory network as target concepts, and the selected target concept is used as the center to divide other concepts in the fully connected semantic memory network according to short distance, medium distance and long distance to form three different concept fields; concepts in the long-distance concept field and the short-distance concept field are combined in pairs to generate more long-distance concept combinations, and the generated concept combinations are screened and retained according to the correlation of the concept combinations, and finally the concepts in the concept combination are sorted according to the semantic distance between the two concepts, so as to provide designers with inspiration and motivation based on concept combinations.

[0023] Furthermore, the method of the second step is as follows:

[0024] (1) Select patent text and perform word segmentation, part-of-speech tagging, and part-of-speech restoration preprocessing operations;

[0025] (2) Determine whether the preprocessed text has a structure that satisfies the main functional expression template. If so, execute step (3); if not, discard the patent;

[0026] (3) Extract the target word corresponding to the main function word of the template, and determine whether its part of speech is a verb. If it is a verb, the target word is the main function verb. If it is not a verb, the target word is the present participle or derived noun form of the main function verb, and execute step (4);

[0027] (4) Search all derivatives of the main functional target word in the semantic dictionary and find the verb form as the main functional verb.

[0028] Furthermore, in the third step, the similarity between the main functional verb and the target functional verb is calculated using semantic similarity based on semantic dictionary path distance.

[0029] Furthermore, in the fourth step (2), a comprehensive semantic similarity calculation is performed based on Word2Vec and WordNet.

[0030] Furthermore, the method for calculating comprehensive semantic similarity based on Word2Vec and WordNet is as follows:

[0031] 1) Select the target functional patent text set, and build a word sequence corpus from the patent texts in the target patent text set to train the Skip-gram model of Word2Vec;

[0032] 2) Select any two design concepts from the target functional patent concept set, obtain the multi-dimensional word vectors of the two concepts through the trained Skip-gram model, and calculate the semantic similarity Sim based on Word2ec wv ;

[0033] 3) Calculate the semantic similarity Sim of the above two concepts based on WordNet path distance wn ;

[0034] 4) Give the semantic similarity Sim wn and Sim wv The weights are assigned and the weighted sum of the two concepts is calculated. The result is the comprehensive semantic similarity of the two concepts. According to the above process, the comprehensive semantic similarity between the concepts in the target functional patent concept set is calculated in sequence.

[0035] Aiming at the discreteness and singularity of massive patent data resource information and the accurate recommendation of patent concept extraction, the present invention proposes a patent semantic memory network model construction method for integrating semantic and domain information. Through machine learning framework, natural language processing and network science methods, the main function template construction of patent data including semantic memory network and patent network, the acquisition of target function patents and the construction of highly creative semantic memory network are realized, so that the network concept combination can be used to assist designers in generating innovative design inspiration incentives. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The overall construction framework of patent semantic memory network

[0037] Figure 2 Common expressions of main functions in patents

[0038] Figure 3 Patent main function verb extraction process based on main function template

[0039] Figure 4 WordNet Hierarchy

[0040] Figure 5 Comprehensive semantic similarity calculation process based on Word2Vec and WordNet

[0041] Figure 6 Functional decomposition tree based on design problem

[0042] Figure 7 Patent statistics chart of main function expression template coverage

[0043] Figure 8 Patented semantic memory network DETAILED DESCRIPTION

[0044] To make the technical solution of the present invention clearer, the present invention is further described below in conjunction with the accompanying drawings. The present invention is specifically implemented according to the following steps:

[0045] 1. Construction of main functional template based on patent knowledge expression

[0046] The main function of a patent is the main purpose achieved by the product or method in the patent, which is usually reflected in the title and abstract of the patent. There are currently two main methods for extracting functional words from patents. One is to parse the SAO (Subject-Action-Object) structure in the patent text through dependency analysis, and use Action and Object as functional verbs and functional nouns respectively; the other is to establish a mapping between the patent text and the functional base, and use supervised learning methods to classify the patent into one or several functions in the functional base. The expression of the main function of the patent in the patent text is usually expressed in the form of verbs, present participles of verbs, and derivatives of verbs, such as Figure 2 As shown in the figure, in dependency analysis, present participles and verb derivative forms are often not analyzed as verbs, making it difficult to correctly extract the SAO structure of the main function. At the same time, there are a large number of SAO structures of non-functional expressions in patent texts. Therefore, the functional word extraction method that parses the SAO structure through dependency analysis is not only difficult to correctly extract the functional words of the main function, but may also extract some non-functional words.

[0047] Although the method of using supervised learning to classify patent text into one or several functions of a functional base avoids analyzing text grammar, it involves supervised learning tasks of multi-classification and multi-label classification. If it is to be ensured that the supervised learning model can accurately map the patent to the functional base of its corresponding main function, a large amount of high-quality patent text data must be annotated to provide the supervised learning model for training. This process requires relevant personnel in various fields to annotate patents in various fields with the same standards to ensure that the entire field of the patent library is covered, which is time-consuming and labor-intensive.

[0048] Aiming at the expression form of the main functions in patents, a method for extracting main functional verbs based on functional expression templates is proposed. Referring to the existing research on the syntax of patent functional expression and combining the expression characteristics of the main functions of patents, seven functional expression templates are proposed as shown in Table 1, involving three types of templates: general, product, and method. Templates 1 and 2 are general functional expression templates proposed with reference to existing research. Templates 3, 4, and 5 are proposed for product invention patents. Typical products are nouns including "device", "equipment", "system", "machine", "mechanism", "apparatus", etc. in the names of invention patents, which can be directly identified as products; specific products refer to nouns with certain specific functional products, mainly verb-derived nouns with "er" or "or" as affixes, such as collector "collector". Templates 6 and 7 are proposed for method invention patents, with "method" as an important marker word.

[0049] Table 1 Patent main function expression template

[0050]

[0051] 2. Extraction of main functional verbs from patent text based on main functional templates

[0052] Based on the main function expression template in Table 1, the main functions expressed in the template in the form of verbs, present participles and derived nouns are extracted and restored to the form of verbs. The specific steps of extraction are as follows: Figure 3 shown.

[0053] (1) Using the patent title and abstract as the original text, perform preprocessing operations such as word segmentation, part-of-speech tagging, and part-of-speech restoration;

[0054] (2) Determine whether the preprocessed text has a structure that satisfies the template in Table 1. If so, execute step (3); if not, discard the patent;

[0055] (3) extracting the target word corresponding to the main function word of the template, and judging whether its part of speech is a verb, if it is a verb, the target word is the main function verb, if not a verb, the target word is the present participle or derived noun form of the main function verb, and executing step (4);

[0056] (4) Search all derivatives of the main function target word in WordNet and find the verb form as the main function verb.

[0057] WordNet is an English dictionary based on cognitive linguistics. It not only contains a large number of English words, but also establishes relationships between words, such as synonymy, hyponymy, etc. Based on WordNet, it is very easy to retrieve all related words of a word.

[0058] Since a patent may satisfy more than one functional expression template, there may be multiple main functional verbs extracted based on the template. For example, "Coating device for spraying objects of different shapes" meets both templates 3 and 4, so the main functional verbs extracted are "coat" and "spray". Both main functional verbs can be used to express the main function of the patent.

[0059] 3. Acquisition of target functional patent text set based on semantic dictionary

[0060] After determining the main functional verbs of each patent, the target functional patent text set for constructing the semantic memory network is obtained according to the similarity between the main functional verbs and the target functional verbs of the patent. The similarity between the main functional verbs and the target functional verbs adopts the semantic similarity based on WordNet path distance.

[0061] In WordNet, nouns, verbs, adjectives, and adverbs are grouped into synsets, each of which contains all words with the same meaning. Figure 4 The hierarchical structure shown is used to express the relationship between synsets. Based on the hierarchical structure of WordNet, the semantic similarity is calculated using Formula 1.

[0062]

[0063] Where w1 and w2 are two words, S1 and S2 are the synonym sets of w1 and w2, respectively, and path_dist(S1, S2) is the shortest path length between the synonym sets S1 and S2 in the hierarchy.

[0064] Formula 1 is used to calculate the WordNet semantic similarity between the main functional verbs of each patent and the target functional verb. If a patent has multiple main functional verbs, the semantic similarity between each main functional verb and the target functional verb is calculated and the maximum value is taken. wn Select Sim wn The threshold is 0.7, and patents with main functional verbs above this threshold are added to the target functional patent text set for constructing the semantic memory network.

[0065] 4. Construction of a highly creative semantic memory network based on semantic computing

[0066] (1) Patent concept extraction based on document word frequency and part of speech

[0067] The concepts in patents are words used to express the structure, function, operation, scenario and other information contained in patents. They are keywords that can express the subject content of patents, so the keywords in patent texts are extracted first. Keyword extraction also uses the title and abstract of patents as the original texts and uses the TF-IDF (Term Frequency & Inverse Documentation Frequency) algorithm for extraction.

[0068] The TF-IDF algorithm is TF×TDF, where TF stands for term frequency, which refers to the frequency of a word appearing in a text. The more important the word, the more times it appears in the text. IDF stands for inverse document frequency, which refers to the document frequency of a specific word appearing in all documents. Words that appear in only a few documents are more important than words that appear in multiple documents. The calculation method of the TF-IDF algorithm is as follows:

[0069]

[0070]

[0071] TFIDF i,j =TF i,j ×IDF i #(4)

[0072] Where n i,j is the number of times word i appears in document j, |D| is the total number of documents, |{j:w i ∈d j} is the number of documents in which word i appears.

[0073] The TF-IDF algorithm can not only highlight the words with high frequency in the document, but also consider the influence of the words that appear frequently in each document, and can better extract keywords from long texts. However, the TF-IDF algorithm still has shortcomings. When obtaining internal information of the text through TF, it simply considers the frequency of the word in the text, and ignores other features such as the position, range, and part of speech; when obtaining external information of the text through IDF, it ignores the category of the word in all documents. If a word often appears in a type of document, then the word can be used as a keyword for this type of document, but the number of this type of documents in the document set far exceeds that of other types of documents, then IDF will weaken the importance of the word.

[0074] In order to avoid the shortcomings of the TF-IDF algorithm as much as possible, the calculation of TF and IDF is modified as follows:

[0075] In the preprocessing stage before calculating TF-IDF, in addition to the conventional preprocessing operations such as word segmentation and stop word removal, part-of-speech tagging is also performed to remove prepositions that have no practical meaning, and word form restoration operations are performed on plural nouns, non-finite verbs, and comparative adjectives and adverbs to avoid the situation where one word has multiple forms as much as possible; the patent title is a condensed summary of the entire patent content, so the words that appear in the title are more important than the words in other positions. In order to reflect their importance, the TF value of the words in the patent title is magnified by 1.5 times; because the patents in the target function patent text set are all patents similar to the target function, the field categories they cover will be relatively concentrated, so IDF will weaken the importance of keywords in this category. This situation can be avoided by expanding the patent text set for calculating IDF to all patents.

[0076] After extracting the keywords, it is necessary to select conceptual words as concepts. The inspiration for product design comes from specific types of words, such as verbs, nouns and other content words. In patent texts, nouns mainly express certain objects, structures, and characteristics, while verbs mainly express certain functions, behaviors, and operations. Adjectives and adverbs are mainly used to semantically supplement the content expressed by nouns and verbs, and it is difficult to express concepts independently. Therefore, the nouns and verbs in the keywords of each patent in the target functional patent text set are used as the conceptual vocabulary of the patent, and also as the nodes of the semantic memory network in the patent semantic memory network.

[0077] (2) Comprehensive concept semantic similarity calculation by integrating semantic dictionary and word embedding

[0078] After extracting all the concept words in the target functional patent text set and determining the nodes in the semantic memory network, the nodes need to be connected with edges to form a complete semantic memory network. The edges in the semantic memory network are the semantic similarity between the two concept words corresponding to the nodes.

[0079] There are many methods for calculating semantic similarity, including dictionary-based, encyclopedia-based, neural network-based, ontology-based, etc., but generally these methods can be divided into two types: semantic dictionary-based and corpus-based. The former is based on artificially constructed semantic dictionaries such as WordNet, Wikipedia, mindnet, etc., and calculates semantic similarity by calculating the distance between words in tree-structured or graph-structured dictionaries; the latter is based on the contextual information of words in large-scale corpora, and calculates semantic similarity by representing words as vectors containing contextual information and calculating the cosine similarity between vectors.

[0080] Semantic similarity based on semantic dictionary is a linguistics-based method. Compared with corpus-based semantic similarity, it does not rely on corpus, but on semantic dictionary compiled by linguistic experts. It does not have noise interference, is not limited to a certain field, and has a wider scope of application. However, for a specific corpus in a specific field, the semantic dictionary-based method is difficult to find the similarity in information between lexical contexts. The corpus-based method can make up for this shortcoming. Therefore, a method combining semantic dictionary-based and corpus-based methods is used to calculate semantic similarity. The semantic dictionary-based semantic similarity adopts the semantic similarity based on WordNet path distance. The calculation method is shown in Formula 1. The corpus-based semantic similarity adopts the semantic similarity based on Word2Vec.

[0081] Word2Vec is a distributed word representation method based on deep learning for word embedding proposed by Google in 2013. It can learn high-quality word vectors containing contextual information from massive corpora. Word2Vec mainly has two models: Continuous Bag-of-Words Model (CBOW) and Continuous Skip-gram Model (Skip-gram). CBOW generates the word vector of the target word by calculating the context of the target word, while Skip-gram generates the word vector of its context based on the target word. Although the two models have different processes for generating word vectors, they essentially calculate and generate multi-dimensional word vectors through the contextual relationship between words.

[0082] As a method of representing words in text, Word2Vec is widely used in the field of natural language processing for tasks such as text classification, entity recognition, and machine translation. The word vectors generated by Word2Vec can also calculate semantic similarity through vector cosine distance, and by comparing with WordNet, it is found that Word2Vec can express richer semantic information than other word representation methods.

[0083] This patent uses the Word2Vec model of the Skip-gram model to generate 300-dimensional word vectors, and the similarity of the words is calculated using the cosine similarity shown in Formula 5.

[0084]

[0085] In the formula, w1 and w2 are two words, and their word vectors are A(a1,a2,a3,…,a 300 ) and B(b1,b2,b3,…,b 300 ).

[0086] The comprehensive semantic similarity calculation process based on Word2Vec and WordNet is as follows Figure 5 As shown in the figure, firstly, a word sequence corpus is constructed from the target patent text set through sentence segmentation, word segmentation, word form restoration, and stop word removal operations to train the Word2Vec Skip-gram model. Then, two concepts are taken from the target functional patent concept set and the semantic similarity Sim based on the WordNet path distance is calculated according to formula 1. wn At the same time, the Skip-gram model is used to obtain the 300-dimensional word vectors of the two concepts, and the semantic similarity Sim based on Word2Vec is calculated according to Formula 5 wv ; Finally, give the semantic similarity Sim wn and Sim wvAssign a weight of 0.5 to calculate the weighted sum of the two concepts, and the result is the comprehensive semantic similarity of the two concepts. According to the above process, the comprehensive semantic similarity between the concepts in the target function patent concept set is calculated in sequence. The obtained comprehensive semantic similarity not only reflects the linguistic similarity of the concepts, but also reflects the similarity of the concepts from the contextual semantic information of the target function patent text set. Among them, the comprehensive semantic similarity of the two concepts is calculated as follows:

[0087] Sim=ω1Sim wn +ω2Sim wv (ω1=ω2=0.5) (6)

[0088] After calculating the comprehensive semantic similarity between all concepts, all concepts can be connected to construct a fully connected concept semantic memory network. The fully connected concept semantic memory network can meet the high creativity conditions to the greatest extent, and any concept only needs to pass through one edge to reach another concept.

[0089] 5. Combination of distant concepts based on semantic distance and relevance

[0090] According to the creative association theory, combining concepts with distant semantic distances has a positive effect on creativity. The greater the semantic distance between the concepts in a new concept combination, the more creative the combination is, and the easier it is to motivate designers to come up with innovative concepts.

[0091] In order to combine concepts with distant semantic distances, several concepts are first selected from the semantic network as central concepts according to the requirements of the design task, and other concepts in the semantic network are divided into three conceptual areas: close distance, medium distance, and far distance based on the selected central concepts, which respectively reflect the different conceptual areas with close, moderate, and far semantic distances from the central concept.

[0092] In order to describe the requirements of the design task as completely as possible, more than one central concept is usually selected. It is necessary to first calculate the semantic distances from other concepts to these central concepts. The calculation method is shown in Formula 7. The average semantic distance from each concept to each central concept is taken as the semantic distance from the concept to the central concept.

[0093]

[0094] In the formula, w i is the concept for which the semantic distance needs to be calculated, Center is the central concept domain, |Center| is the number of central concepts in the central concept domain, min_dist(w i ,w j ) is the concept w i To the central concept j The shortest semantic distance value.

[0095] The above formula can be used to calculate the semantic distance between each concept and the central concept. According to these semantic distances, the concept domains with different distances are divided, and the largest semantic distance concept_dist is taken. max , the semantic distances of all concepts to the central concept are in the interval [0, concept_dist max ], divide the interval into three equal parts, concept_dist near and concept_dist far are two trisection points, as shown in Equations 8 and 9.

[0096]

[0097]

[0098] If the semantic distance from a concept to the central concept is in the interval [0, concept_dist near ], the concept is classified into the close concept field. If it is in the interval [concept_dist near ,concept_dist far ], it is classified into the middle distance concept area. If it is in the interval [concept_dist far ,concept_dist max ], then it is classified into the field of remote concepts.

[0099] Combining concepts in the close concept domain and the distant concept domain in pairs can generate a large number of concept combinations with far semantic distances. Combining concepts in the close concept domain in pairs can generate a large number of concept combinations with close semantic distances to the central concept. Although these concept combinations are not as innovative as those with far semantic distances, they have important reference value. Since the combination process is random, there is no guarantee that the resulting concept combination will be meaningful. Although the creative association theory believes that the farther the semantic distance is, the more innovative the combination is, it also requires that the new combination generated is useful to a certain extent.

[0100] Concept combination cannot rely solely on semantics. Concepts with high semantic similarity may be synonyms or synonyms. Combining such concepts is not very meaningful. If two concepts often appear together in a text, it means that the two concepts are often used together. Such a concept combination is more meaningful. The closer the context distance between the two concepts, the easier it is to understand the concept combination. However, the combination of two concepts that have never appeared in the same text may be meaningless and difficult to understand. In order to ensure that the generated concept combination is useful and meaningful, considering the concept co-occurrence and context distance, the correlation is proposed to measure the comprehensibility of the concept combination. Only when the concepts in the concept combination co-occur, the concept combination has correlation. The concept combination that does not co-occur is directly discarded. The correlation calculation method is shown in Formula 10.

[0101]

[0102] In the formula, w1 and w2 are two concepts in the concept combination, n is the total number of sentences where w1 and w2 co-occur in the target function patent text set, and len(sent i ) is the length of the i-th sentence, min_context_dist(sent i ,w1,w2) is the minimum context distance between w1 and w2 in the i-th sentence.

[0103] The higher the concept combination correlation cor_score is, the more times the concept combination appears, the closer the context distance of the concept combination in the sentence is, and the easier the concept combination is to understand. Select an appropriate threshold for the correlation cor_score, retain concept combinations with correlations higher than the threshold, and sort them according to the semantic distance between the two concepts in the concept combination. Designers can browse these concept combinations based on the semantic distance and select the concept combination they think is appropriate as inspiration.

[0104] 6. Example Application

[0105] (1) Functional analysis based on design issues

[0106] This case solves a design problem of "crushing masonry waste in construction waste" and generates a corresponding design solution. According to the example description, the design problem is functionally analyzed. The total function to be achieved in solving the design problem is to crush masonry waste. The hierarchical decomposition method is used to decompose the total function. The functional decomposition tree is as follows: Figure 6 As shown. The sub-function of breaking bricks and stones is the main function, and the other sub-functions are auxiliary functions. The subsequent concept generation and solution generation are all carried out around the main function.

[0107] (2) Data description

[0108] The data used in this patent are some patent texts disclosed in the Derwent Patent Database in 2020. A total of 104,214 patents were obtained, covering 7 Derwent classification codes, covering the period from January 1, 2020 to December 31, 2020, including patent information such as invention name, application number, abstract, inventor, IPC classification code, claims, instructions, citations, etc.

[0109] The invention nouns, application numbers, abstracts, IPC classification codes, claims, and specifications of these 104,214 patents are saved as fields to form a patent database for subsequent use of the patents.

[0110] To construct a patent semantic memory network, it is necessary to extract the target function patent text set used to construct the patent semantic memory network from the patent database. Seven main function expression templates are used to extract the main functions of all patents in the patent database. The main function expression template proposed in this patent covers the patents in the patent database as follows: Figure 7 As shown in the figure, the sum of the coverage of each template exceeds 100% because the same patent may involve multiple templates. As can be seen from the figure, only 8.79% of the patents are not covered by the template, and the coverage rate exceeds 90%.

[0111] According to the above example description, the target main function is "crush". The semantic similarity calculation method based on WordNet is used to calculate the semantic similarity between the main functions of each patent in the patent database and the target main function "crush". According to the meaning of functional words under different semantic similarities, 0.6 is used as the similarity threshold to select all patents greater than the threshold to form the target function patent text set. The target function patent text set has 3175 patent data involving 11 main functions, including "crush", "pulverize", "grind", "extrude", "smash", "squeeze", "break", "shred", "demolish", "mash" and "fragment". These data will be used to construct the patent semantic memory network in the future.

[0112] (3) Construction of patent semantic memory network

[0113] TF-IDF was used to extract keywords from the titles and abstracts of 3175 patents in the target functional patent text set, and nouns and verbs were taken out as concept words and nodes in the semantic memory network. A total of 1106 concept words were extracted in this step. Then, the title abstracts of all patents in the target functional patent text set were segmented, word segmented, stop words removed, and morphological restoration was performed to construct the corpus for training the Skip-gram model of Word2Vec. After the training, the 1106 concept words were input into the Word2Vec model to obtain the corresponding 1106 300-dimensional vectors, and the cosine similarity was used to calculate the semantic similarity Simwv between the concept words based on Word2Vec. At the same time, WordNet was used to calculate the semantic similarity Simwn between the concept words based on the WordNet path distance, and the weighted sum was calculated to obtain the comprehensive semantic similarity Sim between the concept words. Table 2 lists the comprehensive semantic similarity between some concept words.

[0114] Table 2 Calculation results of comprehensive semantic similarity of concept words

[0115]

[0116] After calculating the comprehensive semantic similarity between all 1,106 concept words, a fully connected patent concept semantic memory network was constructed by taking concepts as nodes and concept relationships as edges.

[0117] According to the example description, four concepts, namely “crush”, “grind”, “brick” and “construct”, were selected as the central concepts from the semantic network. The semantic distance intervals and concept fields of near, medium and far distances were divided for the four central concepts respectively. The division results are shown in Table 3.

[0118] Based on the near, medium and far concept fields of the four central concepts, the total near concept field and the far concept field are divided. The total near concept field is the union of the near concept fields of the four central concepts, with a total of 77 concepts, while the total far concept field is the union of the far concept fields of the four central concepts. Concepts that are already in the total near concept field are no longer included in the total far concept field, with a total of 670 concepts. The near concepts and far concepts are shown in Table 4.

[0119] Table 3 Results of the division of the semantic distance intervals of the central concept

[0120]

[0121] Table 4 Concepts of close distance and long distance

[0122]

[0123]

[0124] The concepts in the total close concept domain and the total long-distance concept domain are combined. The concepts can be combined in pairs to generate a total of 54,516 concept combinations. Then the relevance of the concept combinations is calculated and the relevance threshold is determined to be 30. Only concept combinations with a relevance higher than the threshold are retained, a total of 302 concept combinations, and they are sorted according to semantic distance. The top 10 concept combinations in ascending order of semantic distance are shown in Table 5.

[0125] Table 5 Concept combinations arranged in ascending order of semantic distance

[0126]

[0127] From the 302 concept combinations provided by the semantic network, 22 concepts were selected as concept stimuli to generate 7 design concepts, as shown in Table 6. The combination of concepts such as "crush", "build", "stone", and "concrete" from the concept combinations from sequence numbers 1 to 6 can generate the design concepts "crush stone" and "crushbuild"; the combination of concepts such as "crush", "knife", "rotate", "hammmer", and "multi-stage" from the concept combinations from sequence numbers 7 to 15 can generate the design concepts "crush hammer", "crush hob", "multi-stagecrush", and "crush roll"; the combination of concepts such as "crush", "filter", and "sieve" from the concept combinations from sequence numbers 16 to 22 can generate the design concept "filter sieve". Combining the above 7 design concepts, the description of the design concept is "using hammers, hobs or rollers to perform multi-stage crushing of building stone waste and setting up filter sieves for filtering and screening."

[0128] Table 6 Concept motivation and design concept

[0129]

[0130] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for stimulating innovative design inspiration based on patent semantic network concept combination generation, comprising the following steps: Step 1: According to the target functional verbs of the design requirements, the patent set used to construct the patent semantic memory network is limited, and the main function expression template for the patent text is constructed according to the expression form of the main functions in the patent; Step 2: Extract the main functional verbs from the selected patent data based on the main functional expression template; Step 3: Acquisition of target functional patent text set based on semantic dictionary: After determining the main functional verbs of each patent, the target functional patents for constructing the semantic memory network are acquired according to the semantic similarity between the main functional verbs of the patent and the target functional verbs, and the target functional patent text set is formed; Step 4: Constructing a patent semantic memory network based on semantic computing, the method is as follows: ⑴ Patent concept extraction based on TF-IDF and part of speech. Patent concept refers to the keyword that can express the content of the patent subject. Therefore, the patent texts in the patent target function text set are firstly subjected to keyword concept extraction, and then summarized into a target function patent concept set composed of keyword concepts. The keyword extraction is performed using the TF-IDF algorithm. (2) Calculate the comprehensive semantic similarity based on the method based on semantic dictionary and corpus: After extracting all the concept words in the target functional patent text set and determining the nodes in the semantic memory network, the nodes are connected by edges to form a complete semantic memory network. The edges in the semantic memory network are the semantic similarities between the two concept words corresponding to the nodes; the comprehensive semantic similarity is calculated by the method based on semantic dictionary and corpus; (3) Connect each concept pairwise to construct a fully connected concept semantic memory network; Step 5: Inspiration and motivation of long-distance concept combination. The method is as follows: according to the design task requirements, concepts are selected from the fully connected concept semantic memory network as target concepts, and the selected target concept is used as the center to divide other concepts in the fully connected semantic memory network according to short distance, medium distance and long distance to form three different concept fields; concepts in the long-distance concept field and the short-distance concept field are combined in pairs to generate more long-distance concept combinations, and the generated concept combinations are screened and retained according to the correlation of the concept combinations, and finally the concepts in the concept combination are sorted according to the semantic distance between the two concepts, so as to provide designers with inspiration and motivation based on concept combinations.

2. The innovative design inspiration motivation method based on patent semantic network concept combination generation according to claim 1 is characterized in that: The method for the second step is as follows: (1) Select patent text and perform word segmentation, part-of-speech tagging, and part-of-speech restoration preprocessing operations; (2) Determine whether the preprocessed text has a structure that satisfies the main functional expression template. If so, execute step (3); if not, discard the patent; (3) Extract the target word corresponding to the main function word of the template, and determine whether its part of speech is a verb. If it is a verb, the target word is the main function verb. If it is not a verb, the target word is the present participle or derived noun form of the main function verb, and execute step (4); (4) Search all derivatives of the main functional target word in the semantic dictionary and find the verb form as the main functional verb.

3. The innovative design inspiration motivation method based on patent semantic network concept combination generation according to claim 1 is characterized in that: In the third step, the similarity between the main functional verb and the target functional verb is calculated using the semantic similarity based on the semantic dictionary path distance.

4. The innovative design inspiration motivation method based on patent semantic network concept combination generation according to claim 1 is characterized in that: In the fourth step (2), comprehensive semantic similarity calculation is performed based on Word2Vec and WordNet.

5. The innovative design inspiration motivation method based on patent semantic network concept combination generation according to claim 4 is characterized in that: The method for calculating comprehensive semantic similarity based on Word2Vec and WordNet is as follows: 1) Select the target functional patent text set, and build a word sequence corpus from the patent texts in the target patent text set to train the Skip-gram model of Word2Vec; 2) Select any two design concepts from the target functional patent concept set, obtain the multi-dimensional word vectors of the two concepts through the trained Skip-gram model, and calculate the semantic similarity Sim based on Word2ec wv ; 3) Calculate the semantic similarity Sim of the above two concepts based on WordNet path distance wn ; 4) Give the semantic similarity Sim wn and Sim wv The weights are assigned and the weighted sum of the two concepts is calculated. The result is the comprehensive semantic similarity of the two concepts. According to the above process, the comprehensive semantic similarity between the concepts in the target functional patent concept set is calculated in sequence.

Images