Automatic generation method of patent retrieval formula

Abstract

The invention provides an automatic generation method of a patent retrieval formula, which belongs to the technical field of intellectual property retrieval, and comprises the following steps: acquiring a preliminary keyword set and a retrieval purpose; according to the retrieval purpose, keyword categories are automatically recognized, and the categories comprise technical vocabularies, application scene vocabularies, function vocabularies and exclusion vocabularies; distributing a weight for each keyword by using a weight distribution algorithm; screening the keywords by using a greedy algorithm to generate an optimal keyword combination; the technical vocabularies and the function vocabularies are connected through Boolean logic to form the retrieval formula, synonyms and synonyms are connected through OR, keywords of different types are connected through AND, and antonyms are connected through NOT.

Description

Technical Field

[0001] The present invention relates to the technical field of intellectual property search, and in particular to a method for automatically generating a patent search formula. Background Art

[0002] Patent searches often use field search mode, requiring users to convert search targets into search information, match the correct search fields, and use accurate search formulas for searches. However, patent documents involve a large number of professional terms. The diversity and complexity of these terms increase the difficulty of automatic word segmentation and matching. In addition, the patent database is huge and constantly updated. Patent documents in different technical fields differ in expression and terminology. Cross-domain searches require stronger semantic understanding and generalization capabilities.

[0003] At present, there are certain differences in the data retrieved by different search platforms, and the retrieved data needs to be screened. How to achieve fully automatic search formula generation and generate more accurate results is naturally an urgent problem to be solved. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method for automatically generating patent search formulas. According to the search purpose, the keyword category is automatically identified, a weight allocation algorithm is used to allocate weights to each keyword, and a greedy algorithm is used to screen the keywords to generate an optimal keyword combination; a search formula is generated according to the keyword combination, wherein technical terms and functional terms are connected through Boolean logic to form a search formula, and a variety of search formula combinations are automatically generated to ensure the comprehensiveness and flexibility of the search.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for automatically generating a patent search formula comprises the following steps: Obtain a preliminary set of keywords and search purposes; Automatically identify keyword categories based on search purposes, including technical terms, application scenario terms, functional terms, and exclusion terms; Use a weighting algorithm to assign weights to each keyword; Use the greedy algorithm to filter keywords and generate the optimal keyword combination; Generate a search formula based on the keyword combination, in which technical terms and functional terms are connected through Boolean logic to form a search formula, and synonyms and near-synonyms are connected using "OR", keywords of different categories are connected using "AND", and antonyms are connected using "NOT".

[0006] Preferably, the preliminary keyword set includes technical vocabulary, application scenario vocabulary, functional vocabulary and exclusion vocabulary, and the priority of each category of keywords is dynamically adjusted according to the search purpose.

[0007] Preferably, a word embedding model or a semantic network is used to expand synonyms and near-synonyms for keywords of each category, ensuring that relevant vocabulary is included in the search terms.

[0008] Preferably, the weight allocation algorithm adopts the TF-IDF algorithm, which is as follows: The weight of each keyword is calculated using the formula W(ki)=TF(ki)·IDF(ki) to determine the weight based on the frequency of occurrence of the keyword in technical documents and its distribution in all documents, and the weight adjustment changes dynamically according to the search purpose.

[0009] Preferably, the TF and IDF are obtained in advance by preprocessing the vocabulary in the patent database, and the preprocessing steps are as follows: Construct an inverted index to record each independent word and its frequency of occurrence in patent documents; Use stop word filters to remove common words and reduce unnecessary calculations; Preprocess all documents in the patent database and calculate the TF and IDF of each word; The inverted index and the TF-IDF values ​​of the vocabulary are stored after preprocessing and are regularly updated incrementally according to the updates of the documents in the database.

[0010] Preferably, the execution process of the greedy algorithm includes the following steps: Sort the preliminary keyword set from high to low weight; In each iteration, the keyword with the highest weight is selected, and the relevance of the keyword with other keywords in the candidate keyword set is determined. If the keyword has a low relevance with the keywords in the candidate set, it is added to the candidate set; Repeat the iteration until the preset upper limit of keyword quantity is met or enough categories of keywords are covered.

[0011] Preferably, the relevance of the keywords is calculated by cosine similarity, and the formula is as follows:

[0012] Among them, ki and kj are two keywords. and Its vector representation.

[0013] Preferably, the search formula generation process includes: Arrange the expanded keyword set according to weight; Use Boolean logic to generate the final search query; In order to ensure the comprehensiveness and flexibility of the search results, a variety of search formula combinations are automatically generated, including different levels of synonym expansion or a combination of functional vocabulary and application scenario vocabulary.

[0014] Preferably, the greedy algorithm is optimized based on precision priority or recall priority: When the search purpose is accuracy, give priority to technical core words with high weights and eliminate redundant or irrelevant functional words; When the search purpose is comprehensiveness, more extended words and synonyms are retained to ensure coverage of different application scenarios and functional descriptions.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for automatically generating a patent search formula. According to the search purpose, the method automatically identifies the keyword category, uses a weight allocation algorithm to allocate weights to each keyword, uses a greedy algorithm to screen the keywords, and generates an optimal keyword combination; generates a search formula according to the keyword combination, wherein technical terms and functional terms are connected through Boolean logic to form a search formula, and automatically generates a variety of search formula combinations to ensure the comprehensiveness and flexibility of the search.

[0016] The execution process of the greedy algorithm used in this application will be adjusted according to different retrieval purposes to balance the precision and recall rates. DETAILED DESCRIPTION

[0017] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be used for other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0018] The present invention provides a method for automatically generating a patent search formula, comprising the following steps: Step 1: Obtain a preliminary set of keywords and search purposes; Step 2: Automatically identify keyword categories based on the search purpose, including technical terms, application scenario terms, functional terms, and exclusion terms; Step 3: Use the weight assignment algorithm to assign weights to each keyword; Step 4: Use the greedy algorithm to filter keywords and generate the optimal keyword combination; Step 5: Generate a search formula based on the keyword combination, in which technical terms and functional terms are connected through Boolean logic to form a search formula, and synonyms and near-synonyms are connected using "OR", keywords of different categories are connected using "AND", and antonyms are connected using "NOT".

[0019] In step 1 of the present application, a preliminary keyword set and search purpose are obtained through input and initialization, wherein Input: The user provides a preliminary keyword set K={k1,k2,..,kn} and search purpose (comprehensiveness, accuracy, etc.).

[0020] Initialization: Set weight strategies according to different search purposes. For example, when accuracy is the main focus, keyword selection is more stringent, while when comprehensiveness is the main focus, the keyword set will be expanded as much as possible.

[0021] In step 2, natural language processing (NLP) technology is used to classify the input keywords. Common categories include technical vocabulary, application scenario vocabulary, functional vocabulary, and exclusion vocabulary.

[0022] The classification model can classify keywords based on pre-trained domain semantic models (such as BERT) or industry-specific vocabulary. For example:

[0023] “Battery” is classified as a technical term; “Car” is classified as an application scenario word; “Environmental protection” is classified as a functional word.

[0024] Furthermore, in this application, synonyms and near synonyms are expanded for keywords in each category using word embedding models or semantic networks to ensure that relevant words are included in the search terms. For keywords in each category, synonyms and near synonyms are expanded using word embedding models (such as Word2Vec, FastText) or semantic networks (such as WordNet). Ensure that relevant words are also included in the search terms.

[0025] For example, "battery" can be expanded to "lithium battery" or "storage battery". Expanded formula:

[0026] Sext(ki)={ki}∪{kisyn1,kisyn2,..} Among them, Sext(ki) is the expanded keyword set.

[0027] In step three, the weight assignment algorithm assigns weights and weight adjustments to each keyword, where Weight assignment: Use the TF-IDF algorithm or co-occurrence frequency analysis to assign weights to each keyword. Important keywords will receive higher weights. TF-IDF formula:

[0028] W(ki)=TF(ki)·IDF(ki) Weight adjustment: adjust the weights of keywords of different categories according to the search purpose. Technical terms are usually given higher weights, and functional terms and application scenario terms are dynamically adjusted according to actual needs. The final weight table can determine the keyword priority according to the following formula:

[0029] P(ki)=αW(ki)+βR(ki) Among them, α and β are the adjustment parameters of weight and relevance respectively, and R(ki) is the relevance score between keywords.

[0030] It should be noted that the TF value and IDF value in step 3 need to be obtained in advance by preprocessing the vocabulary in the patent database. The preprocessing steps are as follows: 1. Vocabulary extraction: Extract vocabulary from all documents and clean them up (for example, remove stop words, punctuation, etc.).

[0031] 2. Create an index: Create an inverted index for all the words in the document, that is, record in which documents each word appears and the frequency of its appearance.

[0032] 3. Calculate the IDF value: count the document frequency (DF) of each word, and then calculate its inverse document frequency (IDF).

[0033] The calculation formula for TF value and IDF value is: 1. TF (Term Frequency) – the frequency of occurrence of keywords in technical literature TF (TermFrequency) is used to measure the frequency of a keyword in a single document. Its calculation formula is: ; in: f(ki) is the number of times keyword k appears in a document; ∑jf(kj) is the sum of the occurrence times of all keywords in the document.

[0034] The larger the TF value, the more important the keyword is in the document.

[0035] 2. IDF (Inverse Document Frequency) — the distribution of keywords in all documents IDF (Inverse Document Frequency) is used to measure the importance of a keyword in the entire document collection. The fewer documents a keyword appears in, the higher the IDF value, indicating that the keyword has a higher degree of discrimination. The calculation formula is:

[0036]

[0037] in: N is the total number of documents in the document collection; |d:ki∈d| is the number of documents containing keyword ki (i.e., the number of documents in which keyword ki appears).

[0038] When a keyword appears in many documents, the IDF value will be small, indicating that the keyword has low discrimination (for example, common words such as "the").

[0039] In step 4, keyword screening includes: Automatic screening: Based on weight and co-occurrence analysis, a greedy algorithm is used to gradually select the keywords with the highest weight. Overly redundant or irrelevant keywords will be automatically excluded. For different categories, strategies are applied for screening:

[0040] The technical vocabulary is always retained; The application scenario vocabulary can be partially selected; Functional terms can be dynamically added or excluded based on their impact on the search target.

[0041] Among them, the greedy algorithm is an iterative method that selects the current best local optimal solution at each step, hoping to obtain the global optimal solution through continuous local optimal selection. For keyword screening, the goal of the greedy algorithm is to select appropriate keywords to maximize the retrieval effect (such as recall rate and precision rate) and avoid missed detection or false detection.

[0042] Specifically, The steps of the greedy algorithm are as follows: Step 1: Initialize the candidate set S={}.

[0043] Step 2: Sort the keyword set K from high to low according to the weight W(ki). The higher the weight, the higher the keyword priority.

[0044] Step 3: In each iteration, select the keyword ki with the highest weight that has not yet been added to S, and determine its relevance to other keywords in the current candidate set S.

[0045] If ki and the keywords in S are in different categories or have low correlation, then ki is added to S; If the correlation between ki and the keywords in S is too high, ki is skipped.

[0046] Step 4: Continue to iterate until the following termination conditions are met: Reach the preset upper limit of the number of keywords (for example, 5-7 keywords); or cover enough categories of keywords to ensure comprehensiveness.

[0047] Among them, in Step 3, When selecting keywords, give priority to keywords of different categories to ensure that different technologies, functions, and application scenarios are covered. For example, if the current candidate set S already contains several technical terms, the next step may be to give priority to application scenario terms rather than continue to add more technical terms.

[0048] Before adding a keyword each time, it is necessary to calculate the correlation between the new keyword and other keywords in the candidate set S to avoid excessive repetition or redundancy. For example, use cosine similarity to calculate the similarity between keywords:

[0049]

[0050] If Sim(ki,kj) exceeds the set threshold, it means that ki and kj are too semantically close and should be avoided from being selected at the same time.

[0051] In step 5, the search formula is generated by using Boolean logic: Boolean search formula is automatically generated according to the expanded keyword set and its weight. The following rules are followed:

[0052] Synonyms or extended words are connected using "OR"; Keywords of different categories are connected by "AND"; antonyms or exclusion words are connected by "NOT". Sample search formula:

[0053]

[0054] Among them, k1 and k2 are technical vocabulary and application scenario vocabulary, and k3 is an excluded functional vocabulary.

[0055] Further, the combination generates the search formula: Multiple combination generation: To ensure the comprehensiveness and flexibility of the search, multiple search formula combinations (especially for different categories of keywords) are automatically generated. For example, the following combinations can be generated:

[0056] A combination of technical and functional vocabulary; A combination of technical vocabulary and application scenario vocabulary; Search combination at different levels (different depths of synonym expansion). Example combination:

[0057]

[0058] It should be noted that in this application, the execution process of the greedy algorithm will be adjusted according to different retrieval purposes to balance the precision and recall rates.

[0059] 1. Accuracy first (mainly precision rate): Goal: Reduce false positives and ensure search results are as relevant as possible.

[0060] Strategy: Choose fewer high-weight core keywords, tend to eliminate functional words or application scenario words, and focus on technical core words.

[0061] Greedy algorithm tuning: Increase the weight threshold, select the keywords with the highest weight, and avoid adding redundant words with low weight.

[0062] Improve the filtering criteria for the relevance between keywords and reduce the inclusion of similar words.

[0063] Example: Use only k1 and k2 to form the search formula: k1ANDk2 2. Comprehensiveness first (mainly based on recall rate): Goal: To expand the search scope as much as possible and avoid missing any results.

[0064] Strategy: Keep more extended words and synonyms, and introduce a variety of application scenario words and functional words to ensure the coverage of keywords.

[0065] Greedy algorithm tuning: Lowering the weight threshold allows more keywords to be included even if they have a lower weight.

[0066] Lower the relevance threshold and allow certain similar keywords to exist at the same time to expand the search scope.

[0067] Example: Use extended vocabulary: (k1ORk1syn)AND(k2ORk2syn)ANDk3 Those skilled in the art should understand that the embodiments of the present invention shown in the above description are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

Claims

1. A method for automatically generating a patent search formula, characterized in that: The following steps are involved: Obtain a preliminary set of keywords and search purposes; Automatically identify keyword categories based on search purposes, including technical terms, application scenario terms, functional terms, and exclusion terms; Use a weighting algorithm to assign weights to each keyword; Use the greedy algorithm to filter keywords and generate the optimal keyword combination; Generate a search formula based on the keyword combination, in which technical terms and functional terms are connected through Boolean logic to form a search formula, and synonyms and near-synonyms are connected using "OR", keywords of different categories are connected using "AND", and antonyms are connected using "NOT".

2. The method for automatically generating a patent search formula according to claim 1, characterized in that: The preliminary keyword set includes technical vocabulary, application scenario vocabulary, functional vocabulary and exclusion vocabulary, and the priority of each category of keywords is dynamically adjusted according to the search purpose.

3. A method for automatically generating a patent search formula according to claim 1 or 2, characterized in that: The keywords of each category are expanded into synonyms and near-synonyms using word embedding models or semantic networks to ensure that relevant words are included in the search terms.

4. The method for automatically generating a patent search formula according to claim 1, characterized in that: The weight allocation algorithm adopts the TF-IDF algorithm, which is as follows: The weight of each keyword is calculated using the formula W(ki)=TF(ki)·IDF(ki) to determine the weight based on the frequency of occurrence of the keyword in technical documents and its distribution in all documents, and the weight adjustment changes dynamically according to the search purpose.

5. The method for automatically generating a patent search formula according to claim 4, characterized in that: The TF and IDF are obtained in advance by preprocessing the vocabulary in the patent database, and the preprocessing steps are as follows: Construct an inverted index to record each independent word and its frequency of occurrence in patent documents; Use stop word filters to remove common words and reduce unnecessary calculations; Preprocess all documents in the patent database and calculate the TF and IDF of each word; The inverted index and the TF-IDF values ​​of the vocabulary are stored after preprocessing and are regularly updated incrementally according to the updates of the documents in the database.

6. The method for automatically generating a patent search formula according to claim 1, characterized in that: The execution process of the greedy algorithm includes the following steps: Sort the preliminary keyword set from high to low weight; In each iteration, the keyword with the highest weight is selected, and the relevance of the keyword with other keywords in the candidate keyword set is determined. If the keyword has a low relevance with the keywords in the candidate set, it is added to the candidate set; Repeat the iteration until the preset upper limit of keyword quantity is met or enough categories of keywords are covered.

7. The method for automatically generating a patent search formula according to claim 6, characterized in that: The relevance of the keywords is calculated by cosine similarity, and the formula is as follows: ; Among them, ki and kj are two keywords. and Its vector representation.

8. The method for automatically generating a patent search formula according to claim 3, characterized in that: The search formula generation process includes: Arrange the expanded keyword set according to weight; Use Boolean logic to generate the final search query; In order to ensure the comprehensiveness and flexibility of the search results, a variety of search formula combinations are automatically generated, including different levels of synonym expansion or a combination of functional vocabulary and application scenario vocabulary.

9. The method for automatically generating a patent search formula according to claim 1, characterized in that: The greedy algorithm is optimized based on precision priority or recall priority: When the search purpose is accuracy, give priority to technical core words with high weights and eliminate redundant or irrelevant functional words; When the search purpose is comprehensiveness, more extended words and synonyms are retained to ensure coverage of different application scenarios and functional descriptions.

Images