Patent document similarity comparison system and method

Abstract

This invention provides a document similarity comparison system. Users execute algorithms via computer to output similarity scores, aiming to improve the efficiency and quality of patent examination. The system receives the request content of the patent document to be compared and a large amount of cited document information. Through algorithms, it performs deep semantic analysis and comparison, automatically generating similarity scores and detailed comparison analyses between the request and each cited document. The similarity score evaluation undergoes a two-stage screening: the first stage only compares the existing text for similarity, outputting the citation with the highest similarity; the second stage, after asking the user whether to proceed, fine-tunes the weights of keywords emphasized by the user and recalculates the similarity score. When a high-similarity citation is detected (e.g., a score of 75 or higher), the system can further automatically generate a preliminary draft of the examination opinion, including basic case information, citations, and preliminary reasons for patent rejection. The generation of this examination opinion can provide more predictive, customized suggestions based on the context of the request content, and is highly relevant to the current patent examination context. This invention can significantly shorten the time of traditional manual retrieval and comparison, provide objective and consistent examination assistance, effectively reduce the workload of patent examination, and effectively identify potential relevant citations that are difficult to find through traditional keyword retrieval.

Claims

1. A document similarity comparison system, comprising: An input module is configured to receive the invention title of a patent application pending examination, at least one claim, and at least one citation document. A retrieval module, connected to the input module, is configured to automatically retrieve potential citation documents related to the content of the request from a patent document database. The number of citation documents can be determined by the user, and the retrieval results are provided to the input module. A similarity rating module is configured to: receive the content of the request and information of at least one citation document; and perform semantic analysis and comparison of the content of the request and information of at least one citation document by a computer-implemented algorithm. The semantic analysis and comparison includes weighted analysis of the relative position, components, steps, processes, order, frequency, weight, relevance, and other attributes of words to calculate their importance scores. This outputs a first-stage similarity score and a detailed comparison analysis; and based on the first-stage similarity score, the similarity evaluation module filters out at least one highest similarity citation; and inquires whether to perform a second-stage filtering. If so, it receives at least one emphasized keyword through the input module, and the similarity evaluation module fine-tunes the weight of the emphasized keyword and recalculates a second-stage similarity score; an opinion generation module, connected to the similarity evaluation module, is configured to: filter out the first high-similarity citations or the second high-similarity citations based on the second-stage similarity score (if performed) or the first-stage similarity score; and automatically generate a draft examination opinion based on the invention title, the content of the claim, and the first high-similarity citations or the second high-similarity citations; and an opinion parsing module is configured to receive and analyze a foreign examination opinion document to extract the reasons for rejection and the citation document ID of the foreign examination opinion, and provide them to the opinion generation module as contextual information for the draft examination opinion.

2. The document similarity comparison system as described in Request 1, wherein the cited document information includes a citation number and at least one relevant paragraph of the citation.

3. The document similarity comparison system of Request 1, wherein the algorithm of the similarity rating module uses a Conceptual Association Dataset to perform semantic analysis and compare the cited documents during semantic analysis, and / or the algorithm of the opinion generation module can generate sequence representations of contextual information and candidate predicted responses based on a data-driven learning network.

4. As in Request 1, the document similarity comparison system, wherein the semantic analysis and comparison performed by the similarity rating module further identifies the component composition and / or process integrity between the content of the request and the cited documents, and calculates the internal term prominence (ITP) and external term prominence (ETP) based on the relative position, components, steps, processes, order, frequency, weight, relevance and other attributes of the words in the documents, and calculates the set-specific term prominence (CSTP) as the importance score.

5. As in the document similarity comparison system of Request 1, the opinion generation module can further utilize the principle of context-aware ranking, taking the request, cited content and review history as context, to provide a customized response.

6. A document similarity comparison method, executed by a processor of a computer system, includes the following steps: computer software receiving the invention title of a patent application to be examined, at least one claim content, and at least one citation document information; automatically retrieving potential citation documents related to the claim content from a patent document database, the number of which can be determined by the user, and providing the search results for subsequent processing; a computer-implemented algorithm performing semantic analysis and comparison on the claim content and the at least one citation document information, wherein the semantic analysis and comparison includes weighted analysis of attributes such as relative position of words, components, steps, processes, order, frequency, weight, and relevance of words to calculate their importance scores; thereby outputting a first-stage similarity score and... The process includes: a detailed comparative analysis step; a step of inquiring whether to conduct a second-stage screening, and if so, a step of receiving at least one emphasized keyword, fine-tuning the weight of the emphasized keyword, and recalculating a second-stage similarity score; a step of selecting the first high-similarity citations or the second high-similarity citations based on the second-stage similarity score (if performed) or the first-stage similarity score; a step of automatically generating a draft examination opinion based on the invention title, the content of the claim, and the first high-similarity citations or the second high-similarity citations; and a step of using computer software to receive and analyze a foreign examination opinion document to extract the reasons for rejection and the citation document ID of the foreign examination opinion, and using them as contextual information for the draft examination opinion.

7. The document similarity comparison method as described in Request 6, wherein the cited document information includes a citation number and at least one relevant paragraph.

8. The document similarity comparison method as described in Request 6, wherein the semantic analysis and comparison step utilizes a Conceptual Association Dataset to perform semantic analysis and comparison of the cited documents, and / or the algorithm for the step of automatically generating a draft review opinion can generate contextual information and sequence representations of candidate predicted responses based on a data-driven learning network.

9. The document similarity comparison method as described in Request 6, wherein the semantic analysis and comparison further identify the component composition and / or process integrity between the content of the request and the cited documents, and calculate the internal term prominence (ITP) and external term prominence (ETP) based on the relative position, components, steps, processes, order, frequency, weight, relevance, and other attributes of the words in the documents, and calculate the set-specific term prominence (CSTP) as the importance score accordingly.

10. As in the document similarity comparison method of Request 6, the step of automatically generating a draft of the review opinion can further utilize the principle of context-aware ranking, taking the request, the cited content and the review history as context, to provide a customized response.

Images