A patent search method and system based on semantic understanding
Through patent search methods based on semantic understanding, users' search intentions and domain understanding are analyzed, and key content is selected, which solves the problem that search results in the existing technology do not meet user needs, and achieves efficient and accurate push of patent search results, improving the User experience.
Patent Information
- Application Number
- CN202411331353.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-24
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-24
AI Technical Summary
Existing patent search tools mainly rely on the similarity of keywords and sentences for searching, ignoring the user's needs, resulting in the search results that do not meet the user's intentions, bringing users a bad experience, and the search process is complicated and time-consuming.
Using a patent search method based on semantic understanding, by obtaining the search data input by the user, analyzing the user's search intention, determining whether to search for similarity, obtaining the user's personal information and domain understanding, filtering out key content, and finding similarity to achieve the pre-expectation Set a threshold patent document and generate user search results based on the search layout.
Effectively reduce the search of invalid content, improve user search efficiency and accuracy, provide patent results that are more in line with user needs, reduce the frequency of users viewing invalid files, and improve the intelligence and convenience of patent search.
Smart Images

Figure CN118839040B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of patent retrieval, and in particular to a patent retrieval method and system based on semantic understanding. Background Art
[0002] Patent search is a specific channel and method for finding patents, including the use of various commonly used patent search tools and mastering commonly used search methods. Through patent search, you can understand the status of the existing technology and avoid repeated development, thereby saving research and development time and money. And through the search, you can obtain relevant comparative documents, which will help applicants improve their technical solutions, so as to better propose technical solutions and obtain the best protection effect. At present, there are a large number of search tools for patent search, but the current search tools basically search and judge based on the similarity of key words and sentences to obtain the most similar patents. However, these search tools ignore the needs of users. It is not the patent with the highest similarity to the user's input information that the user wants to obtain. For the searcher, the retrieved patents do not meet personal requirements, which brings a bad experience to the user and requires repeated searches, which makes the search process complicated and time-consuming. Summary of the invention
[0003] The purpose of the present invention is to provide a patent search method and system based on semantic understanding to solve the problems raised in the above background technology.
[0004] In the first aspect, the present application provides a patent search method based on semantic understanding, which adopts the following technical solution:
[0005] Obtain the search data input by the user, and make a preliminary judgment based on the search data to obtain the user's search intention;
[0006] Determine whether to perform a similarity search based on the user's search intent. If a similarity search is performed, obtain the user's personal information.
[0007] Extract the subject area of the search data, analyze the user's understanding of the subject area based on the personal information and the subject area, and obtain the user's understanding of the area;
[0008] Generate technical solutions based on search data, and use domain knowledge to screen technical solutions to obtain key content;
[0009] Search for patent documents whose similarity reaches a preset similarity threshold according to the key content as reference documents;
[0010] Obtain similar positions of reference documents, set similar positions as jump positions, obtain search layouts based on the jump positions, generate patent search results for users based on the search layouts, and push them to users.
[0011] Preferably, the step of obtaining the search data input by the user and preliminarily determining the user's search intention based on the search data is specifically:
[0012] Obtain the search data input by the user, count the length of the search data and record it as the search length;
[0013] Set a critical value for the search length. If the search length exceeds the critical value, the search data is considered to be solution data, and the user's search intention is to find an approximate solution.
[0014] If the search length does not exceed the critical value, the keywords of the search data are extracted;
[0015] Confirm the meaning scope of keywords, extract the collocation of keywords, and determine the user's search intention based on the meaning scope and collocation.
[0016] Preferably, the steps of confirming the meaning scope of the keywords, extracting the collocation of the keywords, and determining the user's search intention according to the meaning scope and the collocation are specifically as follows:
[0017] Obtain hypernyms and hyponyms of the keyword, and form a keyword relationship tree based on all hypernyms and hyponyms;
[0018] According to the keyword relationship tree, a vocabulary level is formed, wherein the level of the hypernym is lower than the level of the hyponym, and the vocabulary level of the keyword is obtained by searching;
[0019] Set a vocabulary level threshold to determine whether the vocabulary level of the keyword reaches the vocabulary level threshold;
[0020] If the vocabulary levels of all keywords reach the vocabulary level threshold, it is determined that the user's search intention is to find similar solutions;
[0021] If there are keywords whose vocabulary levels do not reach the vocabulary level threshold, the matching degree between the keywords is obtained to determine the user's search intention.
[0022] Preferably, if there is a keyword whose vocabulary level does not reach the vocabulary level threshold, the step of obtaining the collocation degree between the keywords to determine the user's search intention is specifically:
[0023] The AD of the keywords is calculated, and the calculation formula is as follows:
[0024] ;
[0025] Among them, AS is the number of patents in which all keywords appear in the same patent, and AP is the frequency of the keywords used in combination during use. is the collocation strength between the keyword and the subject area formed by the keyword, where i is the keyword number and is greater than 0. , , is a proportional coefficient greater than 0, and n is the maximum number of keywords;
[0026] If the matching degree is lower than the preset matching degree threshold, it is judged that the user's search intention is to find similar solutions;
[0027] If the matching degree is not lower than the preset matching degree threshold, it is determined that the user's search intention is to find an existing solution.
[0028] Preferably, the step of determining whether to perform a similarity search based on the user's search intention and obtaining the user's personal information if a similarity search is performed is specifically as follows:
[0029] If the user's search intention is to find an approximate solution, a similarity search is performed;
[0030] If the user's search intention is to find an existing solution, no similarity search is performed, and the subject area is extracted based on the search data;
[0031] Find patent documents that match the subject area as patent search results and push them to users.
[0032] Preferably, the step of extracting the subject area of the search data, analyzing the user's understanding of the subject area based on the personal information and the subject area, and obtaining the user's understanding of the area is specifically:
[0033] The personal information includes education, work industry and patent experience;
[0034] The user's understanding of the subject area is calculated and recorded as the field understanding BL. The calculation formula is as follows:
[0035] ;
[0036] Among them, BX is the correlation between the user's education and the education in the subject field, BG is the correlation between the user's work industry and the work in the subject field, and BJ is the correlation between patent experience and experience in the subject field. , , is a proportionality factor greater than 0.
[0037] Preferably, the BJ is a step of determining the correlation between patent experience and experience in the subject area, specifically:
[0038] Obtain the user's historical search records, and extract the search frequency CS related to the subject area from the historical search records;
[0039] Extract patent documents that match the subject area and record them as subject documents, and search for words with the same meaning as the keywords in the subject documents and record them as synonyms;
[0040] The empirical correlation BJ is calculated, and the calculation formula is as follows:
[0041] ;
[0042] CZ is the proportion of patent work content in the user's work content and is recorded as the patent work proportion. CP is the frequency of occurrence of keywords in synonyms. , , is a proportionality factor greater than 0.
[0043] Preferably, the step of generating a technical solution based on the search data and screening the technical solution using the field understanding to obtain key contents is specifically as follows:
[0044] Generate a technical solution based on the search data, extract the technical points of the technical solution, and determine whether the technical points are applicable to the product;
[0045] If the technology point is applied to a product, the popularity of the applied product is obtained and recorded as the product popularity;
[0046] According to the preset association table of domain knowledge and product popularity, find the product popularity corresponding to the user's domain knowledge and record it as the standard popularity;
[0047] Screen out the technical points whose product popularity reaches the standard popularity and record them as candidate technical points;
[0048] If a technical point is not applied to the product, it will be directly recorded as a candidate technical point. Key technical points are screened from the candidate technical points and recorded as key contents.
[0049] Preferably, the step of selecting key technical points from the candidate technical points and recording them as key contents is specifically as follows:
[0050] Obtain patent documents containing the technical points to be selected and record them as the first patent documents, and calculate the ratio of the first patent documents to all patent documents and record it as the first ratio;
[0051] Determine the patent classification number for searching patents according to the subject area and record it as the search classification number;
[0052] Classify the first patent document according to the patent classification number, extract the first patent document with the same classification number as the search classification number and record it as a similar patent;
[0053] The ratio of similar patents to the first patent is counted and recorded as the classification number ratio, and the weight ratio of the first ratio and the classification number ratio is set respectively. The technology popularization is calculated based on the first ratio, the classification number ratio and the corresponding weight ratio;
[0054] According to the preset association table of field understanding and technology popularization, find the technology popularization corresponding to the user's field understanding and record it as the standard popularization;
[0055] Screen and remove technical points whose technical popularity has reached the standard popularity, obtain the key technical points and use them as the key contents.
[0056] In the second aspect, the present application provides a patent search system based on semantic understanding, which adopts the following technical solutions:
[0057] A patent search system based on semantic understanding, comprising:
[0058] The search intention module obtains the search data input by the user and makes a preliminary judgment on the user's search intention based on the search data;
[0059] The search judgment module determines whether to perform a similarity search based on the user's search intention. If a similarity search is performed, the user's personal information is obtained;
[0060] The domain understanding module extracts the subject domain of the search data, analyzes the user's understanding of the subject domain based on personal information and the subject domain, and obtains the user's domain understanding;
[0061] The key content module generates technical solutions based on the search data and uses the domain understanding to screen the technical solutions to obtain key content;
[0062] The reference document module searches for patent documents whose similarity reaches a preset similarity threshold as reference documents based on key content;
[0063] The search push module obtains the similar position of the reference file, sets the similar position as the jump position, obtains the search layout according to the jump position, generates the user's patent search results according to the search layout, and pushes them to the user.
[0064] In summary, the present application includes at least one of the following beneficial technical effects:
[0065] 1. First, analyze the user's search intent based on the search data entered by the user, and determine whether a similarity search is needed based on the user's search intent. If a similarity search is performed, the user's understanding of the search subject area is obtained based on the user's personal information, and the key content is obtained in combination with the popularity of the solution's technical points. Perform a similarity search on the key content and push it according to the similar positions of the content. This effectively reduces the retrieval of invalid content, improves the user's search efficiency, and improves the convenience of patent retrieval based on semantic understanding.
[0066] 2. Preliminarily judge the user's search intent based on the length of the search data entered by the user. If the length of the search data does not reach the search length critical value, the meaning range of the keyword is determined based on the level of the hyponyms and hyponyms of the keyword in the search data, and the user's search intent is further judged. If the keyword does not reach the vocabulary level threshold, the user's search intent is judged again based on the collocation degree of the keyword. Based on the user's search intent, a preliminary push plan can be obtained, so that the user can retrieve patents that are more in line with their needs, which improves the accuracy of patent searches based on semantic understanding.
[0067] 3. According to the user's education, work industry and patent experience, the user's understanding of the search subject area is analyzed. At the same time, according to the technical points obtained from the user's search data and whether the technical points are applied to the product, the product popularity and technology popularity are obtained respectively. According to the user's understanding of the field, the corresponding popularity is matched, and the key content is screened for similarity search. Searching for the key content required by the user is conducive to reducing the frequency of users viewing invalid patent documents, obtaining the required patent documents more quickly, and improving the intelligence of patent search based on semantic understanding. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] Figure 1 It is a schematic diagram of the specific steps of an embodiment of a patent search method based on semantic understanding of the present invention.
[0069] Figure 2 It is a module connection diagram of an embodiment of a patent search system based on semantic understanding of the present invention. DETAILED DESCRIPTION
[0070] Below is a combination of the embodiments and Figure 1-Figure 2 The present invention is further described in detail, but the embodiments of the present invention are not limited thereto.
[0071] The present invention discloses a patent search method based on semantic understanding, which specifically comprises the following steps:
[0072] Step S1, obtaining the search data input by the user, and preliminarily determining the user's search intention based on the search data.
[0073] Step S2, judging whether to perform a similarity search based on the user's search intention, and if a similarity search is performed, obtaining the user's personal information.
[0074] Step S3, extracting the subject area of the search data, analyzing the user's understanding of the subject area based on the personal information and the subject area, and obtaining the user's field understanding.
[0075] Step S4, generating a technical solution based on the search data, and screening the technical solution using the domain understanding to obtain key content.
[0076] Step S5, searching for patent documents whose similarity reaches a preset similarity threshold according to the key content as reference documents.
[0077] Find patent documents similar to the key content, filter them according to the similarity, and obtain reference documents.
[0078] Step S6, obtain the similar position of the reference document, set the similar position as the jump position, obtain the search layout according to the jump position, generate the user's patent search results according to the search layout, and push them to the user.
[0079] In actual use, when users search for patents, the search results they want to obtain are different due to different search intentions. At the same time, when users search, they do not want to search for all the content. Usually, the innovative content of a technical solution only occupies a part of the technical solution. Therefore, in the search process, for the technical content known to the user, the user does not want to search repeatedly. Therefore, considering the user's search intention and search needs in the search process can reduce the user's screening time, reduce the frequency of users viewing files, find patent documents more quickly, and improve the accuracy and convenience of patent search. Similarly, not all content in a patent document is needed by the user. Locating to a position similar to the user's technical solution can increase the speed of users reading patent documents, obtain more effective search information in the same time, and improve search efficiency. The content position with the highest similarity to the user's technical solution in the reference document is used as a similar position, and the content is displayed on the display page. Each reference document displays the most similar content, forming a layout for push, and locates the content when the user clicks into the details page.
[0080] The steps of obtaining the search data input by the user and preliminarily determining the user's search intention based on the search data are as follows:
[0081] Step S11, obtaining the search data input by the user, counting the length of the search data and recording it as the search length.
[0082] Step S12, setting a critical value for the search length. If the search length exceeds the critical value, it is determined that the search data is solution data, and the user's search intention is to find an approximate solution.
[0083] Step S13: if the search length does not exceed the critical value, the keywords of the search data are extracted.
[0084] Step S14, confirming the meaning scope of the keywords, extracting the collocation of the keywords, and determining the user's search intention based on the meaning scope and the collocation.
[0085] In actual use, during the search process, some users like to enter text, and some users like to enter search terms. The length of the search text entered by different users is inconsistent, that is, the number of text characters is different. Usually, longer text content indicates that the user has a more complete technical solution, and the user's search intention is to find an approximate solution. When the length of the search data entered by the user is shorter, it is necessary to determine whether the user has a technical solution idea. Determining the user's search intention is conducive to providing users with more appropriate and suitable technical solutions according to user needs.
[0086] Confirm the meaning scope of keywords, extract the collocation of keywords, and determine the steps to obtain the user's search intention based on the meaning scope and collocation, specifically:
[0087] Step S141, obtaining hypernyms and hyponyms of the keyword, and forming a keyword relationship tree based on all hypernyms and hyponyms.
[0088] Step S142, forming a vocabulary level according to the keyword relationship tree, wherein the level of a hypernym is lower than the level of a hyponym, and searching to obtain the vocabulary level of the keyword.
[0089] Step S143, setting a vocabulary level threshold, and determining whether the vocabulary level of the keyword reaches the vocabulary level threshold.
[0090] Step S144: If the vocabulary levels of all keywords reach the vocabulary level threshold, it is determined that the user's search intention is to find similar solutions.
[0091] Step S145: If there are keywords whose vocabulary levels do not reach the vocabulary level threshold, the collocation degree between the keywords is obtained to determine the user's search intention.
[0092] In actual use, some users' search data is short, but it does not mean that the users do not have a perfect solution. For example, some users like to use search formulas to search for patents. At this time, although the search content provided by the users is relatively small, the users also have a relatively perfect solution. Similarly, using search formulas to search for patents does not mean that you must have a perfect solution. It is necessary to judge based on the keywords given by the users. The more accurate the keywords are, the greater the possibility of having perfect data. The higher the level of the keywords, the more accurate the meaning of the keywords. For example, data is a relatively large hypernym, and its hyponyms include security data, operation data, transmission data, etc. The hyponyms of security data include encrypted data, access data, etc. If the keywords given by the user are "security", "data", and "processing", then the keyword level is low and the scope is very wide. It is unclear what security data is and how to process it, and it is difficult to get a definite technical solution. At this time, it is necessary to further judge based on the matching degree of the keywords. If the keyword level given by the user is higher, for example, using specific component names such as "P-channel MOS tube", "NPN transistor", "backlight control" as keywords, the lower level is higher, which means that the user has a more complete technical solution for the technical solution that uses these two components at the same time. At this time, the user needs to find an approximate solution of the backlight control type that contains these two components.
[0093] If the vocabulary level of a keyword does not reach the vocabulary level threshold, the steps of obtaining the collocation degree between the keywords to determine the user's search intention are as follows:
[0094] Step S1451, calculate the collocation degree AD of the keywords, the calculation formula is as follows:
[0095] ;
[0096] Among them, AS is the number of patents in which all keywords appear in the same patent, and AP is the frequency of the keywords used in combination during use. is the collocation strength between the keyword and the subject area formed by the keyword, where i is the keyword number and is greater than 0. , , is a proportional coefficient greater than 0, and n is the maximum number of keywords.
[0097] If keywords appear in the same patent document, it means that someone has already generated technical solutions based on these factors. The more keywords appear, the more widespread their use is and the higher the degree of matching. For example, "plants" and "weather" exist in most patent documents in this field because weather affects plant growth. At the same time, some technologies are relatively mature and common, and are often matched in daily life. For example, "data" and "processing", "security" and other words are often matched together, which shows that users are also familiar with this combination. In addition, if the strength of matching between keywords and subject areas is high, it means that they are commonly used technical means in this field and users should be familiar with them, so users have not created new technical solutions. Statistical methods can be used to evaluate the strength of matching between keywords and subject areas, such as mutual information, chi-square test, etc.
[0098] Step S1452: If the matching degree is lower than a preset matching degree threshold, it is determined that the user's search intention is to find similar solutions.
[0099] Step S1453: If the matching degree is not lower than the preset matching degree threshold, it is determined that the user's search intention is to find an existing solution.
[0100] In actual use, the user's keyword lower level is not enough, which does not mean that the user does not have a specific solution. When the collocation degree of some keywords is not high, it means that the user has a relatively unique technical solution, which is different from the current common technical solutions, so the user's search intention is to find an approximate solution. When the keyword collocation degree is high, it means that it is a relatively common collocation, and because the vocabulary level is low and the meaning range is large, it is judged that the user does not have a specific solution in order to understand the existing technology in this field. For example, the keywords given by the user are "energy", "blockchain", and "distributed transactions". The keywords are all relatively high-level words, but energy is not often paired with blockchain and distributed transactions, which means that the user has a relatively novel and unique technical solution. Therefore, it is believed that the user's search intention is to find an approximate solution.
[0101] Whether to perform similarity search is determined based on the user's search intention. If similarity search is performed, the steps for obtaining the user's personal information are as follows:
[0102] Step S21: If the user's search intention is to find an approximate solution, a similarity search is performed.
[0103] Step S22: If the user's search intention is to find an existing solution, no similarity search is performed, and the subject area is extracted based on the search data.
[0104] Step S23, searching for patent documents that match the subject area as patent search results, and pushing them to the user.
[0105] In actual use, if the user's search intention is to find similar solutions, it is necessary to compare the user's solution with the existing patent document solutions, compare the similarity of the solutions, and find the comparison file that is most similar to the user's solution. If the user's search intention is to find an existing solution, that is, to understand the solution situation in this field, the user does not have any solution, and only needs to push patent documents that meet the subject area. For example, the user gives the keywords "pipeline" and "monitoring" at this time, indicating that the user wants to know how the existing pipeline monitoring is achieved. At this time, it is necessary to push patent documents that meet pipeline monitoring, and at the same time reduce the appearance of patents with too high similarity, and push patent documents with large similarity differences and meet the requirements. Since the user wants to understand the existing technology in this field at this time, it is not conducive for the user to understand the existing technology if it finds too many similar patents.
[0106] The steps of extracting the subject area of the search data, analyzing the user's understanding of the subject area according to the personal information and the subject area, and obtaining the user's understanding of the area are specifically as follows:
[0107] Step S31, personal information includes education background, work industry and patent experience.
[0108] Step S32, calculate the user's understanding of the subject area and record it as the area understanding BL, the calculation formula is as follows:
[0109] ;
[0110] Among them, BX is the correlation between the user's education and the education in the subject field, BG is the correlation between the user's work industry and the work in the subject field, and BJ is the correlation between patent experience and experience in the subject field. , , is a proportionality factor greater than 0.
[0111] In actual application, in the process of similarity query, it is necessary to confirm the user's understanding of the searched subject area based on the user's personal information. When users conduct similarity searches, they are confirming that their solutions are creative in order to increase the patent authorization rate. Therefore, in the process of selecting patents, it is necessary to screen the focus of technical solutions based on the user's understanding of the subject area. According to the industrial chain relationship between industries, the integration of industries, and the classification of industries, the degree of correlation between the user's major and the subject area, the degree of correlation between the user's work industry and the subject area, and the degree of correlation between patent experience and the subject area are evaluated. The greater the degree of correlation, the greater the user's understanding of the subject area.
[0112] BJ is the step of the correlation between patent experience and experience in the subject area, specifically:
[0113] Step S321, obtaining the user's historical search records, and extracting the search frequency CS related to the subject area from the historical search records.
[0114] Step S322, extract patent documents that match the subject area and record them as subject documents, search for words in the subject documents that have the same meaning as the keywords and record them as synonyms.
[0115] Step S323, calculate the experience correlation degree BJ, the calculation formula is as follows:
[0116] ;
[0117] CZ is the proportion of patent work content in the user's work content and is recorded as the patent work proportion. CP is the frequency of occurrence of keywords in synonyms. , , is a proportionality factor greater than 0.
[0118] In actual use, some users specialize in the intellectual property industry, while some users come into contact with more patents in their daily work. In this case, they will have a higher understanding of patents and will be exposed to more types and fields of patents. According to the user's search records, we can know which fields of patents the user usually works on. The higher the search frequency related to the subject area, the more familiar the user is with the field, so the higher the correlation between patent experience and the subject area. At the same time, the higher the frequency of keywords with the same meaning, the more patent work content is exposed, and the higher the correlation between patent experience and the subject area. The more patent work content, the more familiar the user is with distinguishing between prior art and non-prior art. For example, some users who have not been exposed to patent business believe that products that have not yet appeared on the market are creative, but users with high patent experience know that some technologies are obvious, and even if there are no existing products, they are still not creative.
[0119] Generate technical solutions based on search data, use domain knowledge to screen technical solutions, and obtain key content steps, specifically:
[0120] Step S41, generating a technical solution based on the search data, extracting technical points of the technical solution, and determining whether the technical points are applicable to the product.
[0121] Step S42: If the technical point is applied to a product, the popularity of the applied product is obtained and recorded as the product popularity.
[0122] Step S43: According to a preset association table of field knowledge and product popularity, the product popularity corresponding to the user's field knowledge is searched and recorded as the standard popularity.
[0123] Step S44, screen out the technical points whose product popularity reaches the standard popularity and record them as the technical points to be selected.
[0124] Step S45: If the technical point is not applied to the product, it is directly recorded as a candidate technical point, and key technical points are screened from the candidate technical points and recorded as key contents.
[0125] In actual application, in order to build a complete solution, patent technology usually involves some existing technologies. If the user knows that this is existing technology, then the user does not want to search for relevant content of the existing technology, but only wants to search for the creative technical points in the solution. Therefore, how to determine the key content for retrieval requires the use of the user's understanding of the field and the popularity of the technical points. If the technical point is applied to a product, the user is analyzed based on the popularity of the product to see whether the user knows that the technical point is an existing technology. For example, user A has a low level of understanding of the medical field, but the popularity of technologies such as stethoscopes and gastroscopes is high, so it is believed that user A knows that it is an existing technology. However, the ultrasonic radiation force deep brain nerve regulation instrument is a relatively advanced technology that uses ultrasound to effectively regulate the deep part of the brain, so it is believed that the user does not understand the technology. If user A's technical solution involves gastroscopy and the use of ultrasonic radiation force to regulate deep brain nerves, the key content is the use of ultrasonic radiation force to regulate deep brain nerves. Finding similar technical solutions will make the user know that the probability of authorization of this technical point is low.
[0126] The steps of selecting key technical points from candidate technical points and recording them as key contents are as follows:
[0127] Step S451, obtaining a patent document containing a technical point to be selected and recording it as a first patent document, and calculating the ratio of the first patent document to all patent documents and recording it as a first ratio.
[0128] Step S452, determine the patent classification number of the searched patent according to the subject area and record it as the search classification number.
[0129] Step S453, classify the first patent document according to the patent classification number, extract the first patent document with the same classification number as the search classification number and record it as a similar patent.
[0130] Step S454, count the ratio of similar patents to the first patent and record it as the classification number ratio, set the weight ratio of the first ratio and the classification number ratio respectively, and calculate the technology popularity based on the first ratio, the classification number ratio and the corresponding weight ratio.
[0131] Step S455: According to a preset association table of field understanding and technology popularization, the technology popularization corresponding to the user's field understanding is searched and recorded as the standard popularization.
[0132] The higher the level of understanding of the field, the more technical content is understood, and the content with lower technical popularity can also be understood. Therefore, for technical content that reaches the standard popularity, it is considered that the user already knows that the content is prior art and there is no need to search it repeatedly.
[0133] Step S456, screening out technical points whose technical popularity reaches the standard popularity, obtaining key technical points and using them as key contents.
[0134] In actual application, some technologies have been patented, but have not been put into use, so users cannot learn about this type of technology in daily life. However, if the technology is used in most patent documents, it means that the technology is relatively mature. The fact that there are many patent documents using this technology under the same classification number as the subject area also shows that this technology is a common technology in this field. For example, the buzzer alarm is a relatively mature technology. Even if the alarm used in the field searched by the user is not a buzzer alarm, the application of this technology is relatively existing. For example, the voice broadcast module is often used in interactive devices. Even if the broadcast function is not used on some interactive devices in banks and hospitals, this technology is still a commonly used technical means in this field, so it is also considered to be prior art. According to the user's understanding of the field, confirm the key content of the user's technical solution for retrieval.
[0135] A patent search system based on semantic understanding, by applying the patent search method based on semantic understanding as described above, comprises:
[0136] The search intention module obtains the search data input by the user and preliminarily determines the user's search intention based on the search data.
[0137] The search judgment module determines whether to perform a similarity search based on the user's search intention. If a similarity search is performed, the user's personal information is obtained.
[0138] The domain understanding module extracts the subject domain of the retrieved data, analyzes the user's understanding of the subject domain based on personal information and the subject domain, and obtains the user's domain understanding.
[0139] The key content module generates technical solutions based on the search data, and uses the domain understanding to screen the technical solutions to obtain the key content.
[0140] The reference document module searches for patent documents whose similarity reaches a preset similarity threshold based on key content as reference documents.
[0141] The search push module obtains the similar position of the reference file, sets the similar position as the jump position, obtains the search layout according to the jump position, generates the user's patent search results according to the search layout, and pushes them to the user.
[0142] In actual use, during the search process, determining the key content that the user wants to search can effectively improve the search efficiency and provide users with a better search experience. For example, when designing the interaction of interactive devices, users take into account language switching, speech speed, etc. These are existing technical points for improving interactive devices. If the search is based on the entire solution, the similarities of the solutions with higher similarity will be in the technology of language switching, speech speed, etc. The innovation that the user wants to search may also be an existing technology. At this time, it will be overwhelmed by a large number of patent documents such as language switching and speech speed, causing inconvenience to the user in applying for authorization. The user's search intention is queried based on the search data, and the technical solution is screened based on the user's understanding of the subject area of the search, and the key content is searched, and the key content is layout, presented and pushed according to the location.
[0143] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.
Claims
1. A patent search method based on semantic understanding, It is characterized in that The following steps are involved: Obtain the search data input by the user, and make a preliminary judgment based on the search data to obtain the user's search intention; The step of obtaining the search data input by the user and preliminarily determining the user's search intention based on the search data is specifically as follows: Obtain the search data input by the user, count the length of the search data and record it as the search length; Set a critical value for the search length. If the search length exceeds the critical value, the search data is considered to be solution data, and the user's search intention is to find an approximate solution. If the search length does not exceed the critical value, the keywords of the search data are extracted; Confirm the meaning scope of keywords, extract the collocation of keywords, and determine the user's search intent based on the meaning scope and collocation; The steps of confirming the meaning scope of the keywords, extracting the collocation of the keywords, and determining the user's search intention according to the meaning scope and the collocation are specifically as follows: Obtain hypernyms and hyponyms of the keyword, and form a keyword relationship tree based on all hypernyms and hyponyms; According to the keyword relationship tree, a vocabulary level is formed, wherein the level of the hypernym is lower than the level of the hyponym, and the vocabulary level of the keyword is obtained by searching; Set a vocabulary level threshold to determine whether the vocabulary level of the keyword reaches the vocabulary level threshold; If the vocabulary levels of all keywords reach the vocabulary level threshold, it is determined that the user's search intention is to find similar solutions; If the vocabulary level of a keyword does not reach the vocabulary level threshold, the collocation degree between the keywords is obtained to determine the user's search intention; If the vocabulary level of a keyword does not reach the vocabulary level threshold, the step of obtaining the collocation degree between the keywords to determine the user's search intention is specifically as follows: The AD of the keywords is calculated, and the calculation formula is as follows: ; Among them, AS is the number of patents in which all keywords appear in the same patent, and AP is the frequency of the keywords used in combination during use. is the collocation strength between the keyword and the subject area formed by the keyword, where i is the keyword number and is greater than 0. , , is a proportional coefficient greater than 0, and n is the maximum number of keywords; If the matching degree is lower than the preset matching degree threshold, it is judged that the user's search intention is to find similar solutions; If the matching degree is not lower than the preset matching degree threshold, it is determined that the user's search intention is to find an existing solution; Determine whether to perform a similarity search based on the user's search intent. If a similarity search is performed, obtain the user's personal information. Extract the subject area of the search data, analyze the user's understanding of the subject area based on the personal information and the subject area, and obtain the user's understanding of the area; Generate technical solutions based on search data, and use domain knowledge to screen technical solutions to obtain key content; Search for patent documents whose similarity reaches a preset similarity threshold according to the key content as reference documents; Obtain similar positions of reference documents, set similar positions as jump positions, obtain search layouts based on the jump positions, generate patent search results for users based on the search layouts, and push them to users.
2. According to the patent search method based on semantic understanding as claimed in claim 1, It is characterized in that The step of determining whether to perform a similarity search based on the user's search intention and obtaining the user's personal information if a similarity search is performed is specifically as follows: If the user's search intention is to find an approximate solution, a similarity search is performed; If the user's search intention is to find an existing solution, no similarity search is performed, and the subject area is extracted based on the search data; Find patent documents that match the subject area as patent search results and push them to users.
3. According to the patent search method based on semantic understanding as described in claim 2, It is characterized in that The step of extracting the subject area of the search data, analyzing the user's understanding of the subject area according to the personal information and the subject area, and obtaining the user's understanding of the area is specifically as follows: The personal information includes education, work industry and patent experience; The user's understanding of the subject area is calculated and recorded as the field understanding BL. The calculation formula is as follows: ; Among them, BX is the correlation between the user's education and the education in the subject field, BG is the correlation between the user's work industry and the work in the subject field, and BJ is the correlation between patent experience and experience in the subject field. , , is a proportionality factor greater than 0.
4. According to the patent search method based on semantic understanding as claimed in claim 3, It is characterized in that The BJ is a step of the correlation between patent experience and the experience in the subject area, specifically: Obtain the user's historical search records, and extract the search frequency CS related to the subject area from the historical search records; Extract patent documents that match the subject area and record them as subject documents, and search for words with the same meaning as the keywords in the subject documents and record them as synonyms; The empirical correlation BJ is calculated, and the calculation formula is as follows: ; CZ is the proportion of patent work content in the user's work content and is recorded as the patent work proportion. CP is the frequency of occurrence of keywords in synonyms. , , is a proportionality factor greater than 0.
5. According to the patent search method based on semantic understanding as claimed in claim 4, It is characterized in that The steps of generating a technical solution based on the search data and screening the technical solution using the domain understanding to obtain key contents are specifically as follows: Generate a technical solution based on the search data, extract the technical points of the technical solution, and determine whether the technical points are applicable to the product; If the technology point is applied to a product, the popularity of the applied product is obtained and recorded as the product popularity; According to the preset association table of domain knowledge and product popularity, find the product popularity corresponding to the user's domain knowledge and record it as the standard popularity; Screen out the technical points whose product popularity reaches the standard popularity and record them as candidate technical points; If a technical point is not applied to the product, it will be directly recorded as a candidate technical point. Key technical points are screened from the candidate technical points and recorded as key contents.
6. According to the patent search method based on semantic understanding as claimed in claim 5, It is characterized in that The step of selecting key technical points from the candidate technical points and recording them as key contents is specifically as follows: Obtain patent documents containing the technical points to be selected and record them as the first patent documents, and calculate the ratio of the first patent documents to all patent documents and record it as the first ratio; Determine the patent classification number for searching patents according to the subject area and record it as the search classification number; Classify the first patent document according to the patent classification number, extract the first patent document with the same classification number as the search classification number and record it as a similar patent; The ratio of similar patents to the first patent is counted and recorded as the classification number ratio, and the weight ratio of the first ratio and the classification number ratio is set respectively. The technology popularization is calculated based on the first ratio, the classification number ratio and the corresponding weight ratio; According to the preset association table of field understanding and technology popularization, find the technology popularization corresponding to the user's field understanding and record it as the standard popularization; Screen and remove technical points whose technical popularity has reached the standard popularity, obtain the key technical points and use them as the key contents.
7. A patent search system based on semantic understanding, It is characterized in that By applying a patent search method based on semantic understanding as described in any one of claims 1 to 6, comprising: The search intention module obtains the search data input by the user and makes a preliminary judgment on the user's search intention based on the search data; The search judgment module determines whether to perform a similarity search based on the user's search intention. If a similarity search is performed, the user's personal information is obtained; The domain understanding module extracts the subject domain of the search data, analyzes the user's understanding of the subject domain based on personal information and the subject domain, and obtains the user's domain understanding; The key content module generates technical solutions based on the search data and uses the domain understanding to screen the technical solutions to obtain key content; The reference document module searches for patent documents whose similarity reaches a preset similarity threshold based on key content as reference documents; The search push module obtains the similar position of the reference file, sets the similar position as the jump position, obtains the search layout according to the jump position, generates the user's patent search results according to the search layout, and pushes them to the user.
Citation Information
Patent Citations
Knowledge graph-based question and answer intention identification method
CN111291156A
Semantic information retrieval method based on big data
CN116756347A
