200939045 九、發明說明： 【發明所屬之技術領域】 本發明是關於一種文件比對的方法，特別是一種有關 於專利文件的比對方式。 η. 、 【先前技術】 專利檢索在研發過程中佔有非常重要的地位。倘能在 研發之初進行專利技術的檢索，一來可以透過檢索專利技 〇 術暸解目前技術與競爭者的分佈狀況，甚者透過專利授權 及技術移轉以gp省自行研發的成本；二來可以避免事後因 為侵權訴訟而導致公司的無形和有形資產損失。 隨著科技的演進，以及越來越多企業和學界等研究機 構對於專利的重視，專利不但在數量上快速地成長，專利 技術也越來越複雜。例如美國專利商標局自1994年開始， 每年，專巾％案已超過^ Q萬件，而每年的專利核准量超 過5萬件。中華民國的專利自1997年起，每年申請量超過 5 ^件’核准量每年超過2萬5计件。而在科技的發展上， 以不米技術為例，所涵蓋的領域包含物理、化學、材料及 •生物醫學等，而歐洲專利局甚至新設一個歐洲專利分類號 Y用來追縱和記錄與奈米技術相關的專利。 目月’』在專利檢索上，最普遍採用的方式是透過關鍵字 以及邏輯運算語法，在不同的專利攔位以進行比對分析。 此外’在專利編號第1268437號和專利編號第1249110號 專利文獻中’分别指出利用文字探勘、斷詞的技術，可將 5 200939045 =的特徵值或是文字關連性取出並進行比對。採用上述 的專利數量還是很Α’仍需要仰賴技術人員逐篇 地閱續專利權全文以篩選相關的專利。 Ο 一再者’專利的文件與-般文件所不同之處在於’第 嫌由於專利文件多含有發明元件㈣構，而發明元件的 =可能是機械組合關係、原料合成關係歧方法步驟， =利用關鍵字去檢索帅無法正確表達發明元件架構的關 ^。第二’在進行專利前案檢索或是侵權比對分析時，最 要地是正確解射請專職圍；但若料利相關專業人 士無法準確地㈣專利的保護範圍，且即便利用關鍵字檢 索’亦也無法真實正轉抓住申請專利範圍。 【發明内容】 針對前述的問題，本發明可自動解析申請專利範圍的 文字’並轉化文字成為樹狀的結構圖，故一般不懂申請專 利範圍文法的人也可以透過結構圖以輕易地瞭解技術的架 〇 構。再者，申請專利範圍是專利技術所核心主張的部份， 故可以申請專利範圍為中心，進行檢索前案技術，比較能 ，找到跟技術直接相關的專利。 本發明之第一構想，在於提供一種專利文件比對的方 ' 法，用於比對第一專利文件與第二專利文件，其中第一專 利文件與第二專利文件分別包含一組第一詞彙與一組第二 詞囊。 根據上述構想，本發明之專利文件比對的方法包含下 6 200939045 列步驟’姻第-詞彙建立第—語意架構；第二詞棄 建立第二語意架構；崎第—詞彙與第二詞彙或第-語意 架1與第二語意架構。其中，第一詞彙與第二詞彙為專利 文件中可以定義發明元件的詞彙妓與發明元件相 詞彙。並彻詞彙間的語義連結_，建立語意架構，透 過第-詞彙與第二詞彙的相似度計算或是第—語意 與第二語4架構的相似度，以計算得出文件比對的結果。 Ο ❹ 、根據上述構想’第一詞彙可以從第一專利文件 項或是說㈣内容_’第二詞彙可以從第二專利 請求項或是㈣書内容_，請求歡蚊祕明元件以 及其X件間的關係，而說明書内容則擴充發明元件的 詞彙。 根據上述構想’第-詞彙與第二詞彙透過詞語標 統（POS)分析專敎件的請求項與說财内容的句型，並 擷取發明文件與發明文件相關詞彙。 、 根據上述構想，第-語意架構與第二語意架構透過常 規表示式（Regular expression)擷取詞彙間的關係，並依詞 彙間的關係建立該第一語意架構與該第二語意架構。 根據上述構想，第一語意架構與第二語意架構包含 一從屬關係（Comprising relation) ’此關係描述元件間3 成關係。 、' /根據上述構想，依據不同的重要性，給予第一從屬關 係與第二從屬關係不同的權重以進行比對。 根據上述構想，第一語意架構與第二語意架構包含 7 200939045 一動詞關係（Verb relation) ’此關係描述元件間存在的動 作。 根據上述構想’依據不同的重要性給予第一動詞關係 與第二動詞關係不同的權重以進行比對。 ' 根據上述構想，第一語意架構與第二語意架構包含一 、 連接關係（Preposition relation) ’此關係描述元件間連結的 關係。 根據上述構想，依據不同的重要性給予第一連結關 〇 係與第二連結關係不同的權重以進行比對。 根據上述構想，第一語意架構與第二語意架構包含一 對等關係（Linking relation)，此關係描述元件間可以互相 置換的關係。 根據上述構想，依據不同的重要性給予第一對等關係 與第二對等關係不同的權重以進行比對。 根據上述構想，依據不同的重要性給予第一詞彙與 第二詞彙不同的權重以進行比對。 本發明之第二構想在於提供一種專利文件比對的方 法，用於比對第一專利文件與第二專利文件，其中第一專 利文件與第二專利文件分別包含一組第一詞彙與一組第 二詞彙。 本發明第二構想之專利文件比對的方法包含下列步 驟利用第-詞彙建立一第一語意架構；利用第二詞囊建 =。第一语意架構；比對第一詞彙與第二詞彙以及第一語 &架構與第二語意架構。其中，第—詞彙與第二詞囊為專 200939045 利文件中可以定義發明元件的 的詞彙，並利用該组疋與發Κ牛相關 意架構，透』義連結的關係建立出與語 纽音架禮盥笛°彙一第一岡彙的相似度計算或是第一 ii。、—語意架構的相似度，計算得出文件比對的 _本發明第三構想在於一種儲存媒介，可儲存第一及第 Ο 該儲存媒介可以是硬碟機、快閃記憶 體先碟儲存模組等習知之儲存媒介。 、本案得藉由下列圖式及詳細說明，俾得對本發明之技 述内谷與詳細步驟有更深入了解。 【實施方式】 本發明提出一種專利文件比對的系統方法架構，請 參閱第一圖，其係一系統流程圖，用以說明根據本發明之 一較佳實施例之專利文件比對的系統方法架構。 本發明系統的架構如第一圖所示，總共有五個步 ❹驟，其步驟分別為檢索潛在專利前案技術(patent prior art search) (102)、建構專利語意架構（claim construction) (111)、§吾意架構詞彙的擴充(Claim alignment)、語意架構 ' 連線關係的分析（Link analysis) (105)以及篩選相似的專利 ' 鈿案技術(The finding of relevant prior art) (106)。首先， 使用者先提出第一專利(User’s patent) (101)，並利用該第 一專利找出相近專利前案技術。在第一步驟『檢索潛在專 利前案技術』的模組中，可以利用關鍵字或者是專利分類 9 200939045 號進行初步的檢索，通常這—賴檢索“專利量合很 大，但是可以協助做專利『檢全』的動作，也就是可二確 保不會漏失任何有可能成為前案技術的專利。^一步驟 中，所檢索完的專利資料會變成系統的潛在專利前案技術 (candidate prior art) (107)，即第二專利，而這也 ^剎:200939045 IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to a method of comparing files, and more particularly to a comparison method relating to patent documents. η., [Prior Art] Patent search plays a very important role in the research and development process. If the patent technology can be searched at the beginning of R&D, the patent technology can be searched to understand the current distribution of technology and competitors, and even through the patent licensing and technology transfer, the cost of self-developed by gp province; It is possible to avoid the loss of the company's intangible and tangible assets after the event due to infringement lawsuits. With the evolution of technology and the increasing emphasis on patents by research institutions such as companies and academics, patents have not only grown rapidly in number, but patent technology has become more complex. For example, since the United States Patent and Trademark Office began in 1994, the number of special cases has exceeded 20,000 per year, and the number of patents approved each year has exceeded 50,000. Since the Republic of China’s patents have been filed more than 5^ per year since 1997, the number of approvals exceeds 25,000 per year. In the development of science and technology, taking the rice technology as an example, the fields covered include physics, chemistry, materials and biomedicine, and the European Patent Office even has a new European patent classification number Y for tracking and recording with Nai. Patent related to rice technology. In the patent search, the most common way is to use the keyword and logical operation syntax to perform comparison analysis on different patent blocks. In addition, in the patent documents No. 1268437 and Patent No. 1249110, respectively, the techniques of word exploration and word breaking are respectively indicated, and the characteristic values or textual correlations of 5 200939045 = can be taken out and compared. The number of patents mentioned above is still very low. It is still necessary to rely on the technicians to read the full text of the patents one by one to screen the relevant patents. Ο Repeatedly, the patent document differs from the general document in that it is suspected that the patent document contains more invented components (four), and that the invented component = may be a mechanical combination relationship, a raw material synthesis relationship method step, = utilization key The word to search handsome can not correctly express the structure of the invention component. The second 'in the case of pre-patent search or infringement comparison analysis, the most important thing is to correctly dismiss the full-time job; but if the relevant professionals can not accurately (4) the scope of patent protection, and even use keyword search 'It is also impossible to truly grasp the scope of patent application. SUMMARY OF THE INVENTION In view of the foregoing problems, the present invention can automatically analyze the text of the patent application scope and convert the text into a tree-like structure diagram, so those who generally do not understand the patent application scope grammar can also easily understand the technology through the structure diagram. Architecture. Furthermore, the scope of application for patents is part of the core claim of the patent technology, so it is possible to apply for the patent scope as the center, to carry out the pre-search technology, and to find patents directly related to technology. The first concept of the present invention is to provide a method for comparing patent documents for comparing a first patent document with a second patent document, wherein the first patent document and the second patent document respectively comprise a set of first words. With a second set of words. According to the above concept, the method for comparing patent documents of the present invention includes the following steps: the following steps: 2009, 2009, 45, syllabus, vocabulary establishment, semantic structure, second word, establishment of second semantic structure, and the second vocabulary and second vocabulary or - semantic framework 1 and second semantic architecture. Among them, the first vocabulary and the second vocabulary are words in the patent document that can define the vocabulary of the inventive component and the vocabulary of the inventive component. And the semantic link between vocabulary _, establish a semantic structure, through the similarity calculation of the first-vocabulary and the second vocabulary or the similarity between the first-language meaning and the second-language 4 architecture to calculate the result of the file comparison. Ο 、 According to the above concept, the first vocabulary can be requested from the first patent document or (4) content _ 'the second vocabulary can be requested from the second patent request or (4) the book content _, requesting the mosquito element and its X The relationship between the pieces, while the content of the specification expands the vocabulary of the inventive elements. According to the above concept, the first vocabulary and the second vocabulary analyze the sentence type of the request item and the financial content through the word quotation (POS), and draw the vocabulary related to the invention file and the invention file. According to the above concept, the first semantic structure and the second semantic structure learn the relationship between the vocabulary through a regular expression, and establish the first semantic structure and the second semantic structure according to the relationship between the vocabulary. According to the above concept, the first semantic structure and the second semantic structure include a Comprising relation 'this relationship describes the relationship between the components. According to the above concept, the first subordinate relationship and the second subordinate relationship are given different weights for comparison according to different importance. According to the above concept, the first semantic structure and the second semantic structure include 7 200939045 verb relationship (Verb relation) 'this relationship describes the action existing between the elements. According to the above concept, the weights of the first verb relationship and the second verb relationship are given different weights according to different importance for comparison. According to the above concept, the first semantic structure and the second semantic structure include a "preposition relation" which describes the relationship between the elements. According to the above concept, the first link and the second link are given different weights for comparison according to different importance. According to the above concept, the first semantic structure and the second semantic structure comprise a linking relation, which describes the relationship between components that can be replaced with each other. According to the above concept, the first peer-to-peer relationship is given a different weight than the second peer-to-peer relationship for comparison according to different importance. According to the above concept, the first vocabulary is given a different weight from the second vocabulary for comparison according to different importance. A second concept of the present invention is to provide a method for comparing patent documents for comparing a first patent document with a second patent document, wherein the first patent document and the second patent document respectively comprise a set of first words and a set of The second word. The method of comparing the patent documents of the second concept of the present invention comprises the steps of: constructing a first semantic structure using the first vocabulary; and constructing = using the second word. The first semantic structure; the first vocabulary and the second vocabulary; and the first language & architecture and the second semantic architecture. Among them, the first vocabulary and the second vocabulary are the words that can define the components of the invention in the 200939045 document, and use the relationship between the group and the yak-related structure to establish a relationship with the syllable link. The similarity calculation of the first ceremony of the ceremony is the first ii. - the similarity of the semantic structure, the calculation of the file comparison - the third concept of the present invention is a storage medium, which can store the first and the third storage medium can be a hard disk drive, a flash memory first disk storage mode Groups and other known storage media. In the present case, the following drawings and detailed descriptions can be used to gain a deeper understanding of the technical description and detailed steps of the present invention. [Embodiment] The present invention provides a system method architecture for patent file comparison. Please refer to the first figure, which is a system flowchart for explaining a system method for comparing patent documents according to a preferred embodiment of the present invention. Architecture. The architecture of the system of the present invention is as shown in the first figure. There are a total of five steps, the steps of which are to retrieve the patent prior art search (102) and construct the patent claim construction (111). ), § I intend to expand the vocabulary (Claim alignment), semantic structure 'Link analysis' (105) and screen the similar patent 'The finding of relevant prior art' (106). First, the user first submits a User's patent (101) and uses the first patent to find a similar patented technology. In the first step of the module "Searching for Potential Patent Pre-Technology", you can use the keyword or patent classification 9 200939045 for preliminary search. Usually, this is a large amount of patents, but it can help to make patents. The action of "checking the whole" is to ensure that no patents that might become the technology of the previous case will be missed. In one step, the retrieved patent data will become the system's potential patent prior art. (107), the second patent, and this is also a brake:
入『建構專利語意架構』模組。在『建構專利語意架構』 模組中，系統會分析請求項的語意，並從請求項中找出與 發明元件（components)相關的詞彙，並依據發明元件的^ 結關係展開一 §吾意架構(Claim hierarchy)，而語音架構合 以樹狀結構來呈現其發明元件的關係。 曰 在『語意架構詞彙的擴充』模組中，透過專利說明書 的内谷來擴充語意架構的詞彙。一般請求項的文字通常是 屬於上位詞用法，例如氧氣或二氧化碳，在請求項可能只 用氣體來描述。但系統利用相似度比較，例如c〇sine similarity在專利說明書中找出語意架構具體實施例的描 述’並透過描述找出與語意架構相關且具體的詞彙。而依 據使用者所選定的該第一專利而產生的語意架構，本發明 中稱為『目標語意架構』（target claim hierarchy) (1〇8)，而 經由潛在專利前案技術所產生的語意架構，在此稱為『前 案語意架構』（prior art claim hierarchy) (103)。依據目標 語意架構所進行的語意詞彙擴充，在此稱為目標語意架構 擴充(Target claim alignment) (109)，依據前案語意架構所 進行的詞彙擴充，在此稱之為前案語意架構詞彙擴充 (prior art claim alignment) (104) ° 200939045 在『语意架構連線關係分析』的模組當中，系統會依 據^元件不同的連線關係給予不同的權重，並比對目標 叩思架構與别案語意架構的元件連線關係的相似程度。在 $索相似的專利前案技術』馳巾，线會依據相似程 度篩選出結構相似的專利（丨丨〇)。 、 在建構專利語意架構』的模組當中，系統會分析請 求項的句型，並抓出發明元件的詞彙，進而依照發明元件 詞彙的關係建構-語意架構。系統首先利用stanford所發 〇 展出來的parser對請求項進行文法的分析並抓出名詞或 名祠片語，而這些名詞或名詞片語通常為發明的元件的主 要詞彙。在分析發明元件的詞彙時，將請求項書寫的限制 加入分析規則中，例如，在請求項中第一次出現的發明元 件刖面必須加上不定冠詞『a』、『an』或是用複數形式『a plurality of』來指定之。如果此發明元件在之前有被提過， 當匕第一次出現的時候就必須在前面加上定冠詞『the』 或是『said』’表示這元件跟前面提的元件是同一個。套用 ❹ 這些文法的線索可以協助系統更精準地擷取發明元件。 在抓出發明元件後，接下來會用regular expression 去分析元件詞彙間的關係。除了透過一般的書寫文法給予 的提示外，還可以將請求項特有的文法結構納入考量，例 ·- 如專利通常有前言（Preamble)、連接詞（Transitional phrase) 以及主體（Body)。混合式（Combination type)寫法為『A chair [Preamble] comprising [Transitional phrase]: a lag, and a back [Body]』；而吉普森式(Jepson type)的寫法通常 11 200939045 會先將前案技術寫出，並在後面用『the improvement comprising』或者是『characterized in that (by)』標出技 術的特點’例如『A chair comprising a leg, and a back [Preamble], characterised in that [Transitional phrase] a v wheel on the leg [Body]』° ' 根據請求項文法的結構，總共分析四種的關係，其分 另1J為從屬關係（Comprising relation)、動詞關係（Verb relation)、對等關係（Unking relation)以及連接詞關係 O (Preposition relation)。在此模組中，利用第二圖語意架構 分析圖來解釋此四種關係，第二圖是分別是從專利 7137803『Apparatus for processing a substrate comprising: a moldable layer disposed on a surface of the substrate; a mold with a molding surface comprising a pattern』以及專 利 5772905『A lithographic method for forming a pattern in a film carried on a substrate, comprising the steps of: depositing a film on the substrate; obtaining a mold of a © stiff material which is hard relative to the film, the mold having a first protruding feature』所展開的一第一專利語意 架構（21)及一第二專利語意架構（22)。 其中，第二圖中每一個圓形點表示一個發明元件， A0 (211)代表 Apparatus，A1 (212)代表 substrate，A2 (213)代表 moldable layer’ A3 (214)代表 m〇ld，A4 (215) 代表 surface，A5 (216)代表 molding surface 以及 A6 (217)代表 pattern ;而 B0 (221)代表 lithographic 12 200939045 method，B1 (222)代表 pattern，B2 (223)代表 mold， B3 (224)代表 film 以及 B4 (225)代表 first protruding feature。第二圖中的實線箭頭（218)表示從屬關係，例如在 第二圖中專利7137803依據其在請求項的敘述，發明元件 * A1 (substrate)是由發明元件 A2(m〇ldable layer)和 ** A3(mold)所構成的’而這種從屬關係就用實線箭頭來表 示。而虛線(219)表示連接詞的關係，例如圖中發明元件 A3(mold)和 A5(molding surface)中間有用『with』去連 〇 接這兩個發明元件。如果兩個發明元件中間有動詞連結， 通常是合成或是製造的方法時，使用虛線(226)來表示動 詞的關係’例如用發明元件AO(Apparatus)和A1 (substrate) 之間有製造（processing)的關係，而發明元件 BO(lithographic method)和 B1 (pattern)有合成（forming)的 關係。最後’在撰寫請求項時通常會描述許多態樣，這些 態樣彼此間是可以取代的，以實線(23)表示兩個發明元件 是屬於對等的關係。然而系統在處理此種對等的關係時， ❹ 會把對等關係的發明元件合併在一個節點中處理，表示兩 者發明元件實質上相同，是可以互相替換的。 在建構語意架構的時候，每一個節點都以請求項發明 元件的詞彙來描述，而發明元件的詞彙通常是屬與比較上 ·- 位的用法。以在此模組中，系統搭配專利說明書的内容來 擴充每一個節點的語意以協助之後的分析和比對。故先建 立一個Stopword list榕案把說明書内容不重要的詞彙先行 過濾掉’再者’利用Porter Stemming去除詞語字尾的變 13 200939045 化，例如進行式的『ing』或是過去式的『ed』。如此，進 行詞彙統計的時候就不會因為同一字，卻因為不同型態而 被當作不同字做處理。 經過Stopword以及Stemming的處理後，系統接下來 利用TF(Term Frequency)計算在說明書内容出現頻率較高 的字’並取一個門檻值作為篩選出來的標準。最後，說明 書的詞彙會跟語意架構裡面的詞彙一起計算mutuai information，如下面的公式：Enter the "Building Patent Linguistic Architecture" module. In the "Building Patent Linguistic Architecture" module, the system analyzes the semantics of the request items, and finds the vocabulary related to the invention components from the request items, and develops a § structure according to the relationship of the invention components. (Claim hierarchy), and the speech architecture is combined with a tree structure to present the relationship of its inventive components.曰 In the “Expansion of the semantic structure vocabulary” module, the vocabulary of the semantic structure is expanded through the inner valley of the patent specification. The text of a general request item is usually a subordinate word usage, such as oxygen or carbon dioxide, and the request item may be described only with gas. However, the system utilizes similarity comparisons, such as c〇sine similarity, to find a description of a specific embodiment of the semantic architecture in the patent specification' and to find a specific vocabulary related to the semantic structure through the description. The semantic structure generated according to the first patent selected by the user is referred to as "target claim hierarchy" (1〇8) in the present invention, and the semantic structure generated by the potential patent pre-existing technology. This is referred to herein as the "prior art claim hierarchy" (103). The semantic lexical expansion according to the target semantic structure is referred to herein as the target claim alignment (109). The vocabulary expansion based on the pre-sentence semantic structure is referred to herein as the vocabulary expansion of the pre-case semantic structure. (prior art claim alignment) (104) ° 200939045 In the module of "Spiritual Architecture Connection Analysis", the system will give different weights according to the different connection relationships of the components, and compare the target architecture and other cases. The similarity of the component connection relationship of the semantic architecture. In the similar patented patent technology, the line will screen out similar patents (丨丨〇) according to the similarity degree. In the module of constructing the patent semantic structure, the system analyzes the sentence pattern of the request item, and grabs the vocabulary of the invention component, and then constructs the semantic structure according to the relationship between the vocabulary of the invention component. The system first uses the parser developed by Stanford to analyze the grammar of the request item and grab the noun or the noun phrase, which are usually the main words of the invented component. When analyzing the vocabulary of the inventive component, the restriction of the writing of the request item is added to the analysis rule. For example, the first occurrence of the invention component in the request item must be preceded by the indefinite article "a", "an" or plural. The form "a plurality of" is specified. If the inventive component has been mentioned before, the first word "the" or "said" must be preceded by the definite article "the" or "said" to indicate that the component is the same as the previously mentioned component. Applying 线索 These grammar clues can help the system capture the inventive components more accurately. After the invention component is captured, the regular expression is used to analyze the relationship between the component vocabulary. In addition to the prompts given by the general written grammar, the grammatical structure specific to the request item can also be considered. For example, the patent usually has a preamble, a transitional phrase, and a body. The combination type is "A chair [Preamble] includes [Transitional phrase]: a lag, and a back [Body]"; and the Jepson type is usually written in 2009 20094545. Write, and use "the improvement includes" or "characterized in that (by)" to mark the characteristics of the technology. For example, "A chair includes a leg, and a back [Preamble], characterizedd in that [Transitional phrase] Av wheel on the leg [Body]』° ' According to the structure of the request grammar, a total of four relationships are analyzed, which are divided into 1 Dependency (Comprising relation), Verb relation (Verb relation), and Peer relationship (Unking relation). ) and the conjunction relationship O (Preposition relation). In this module, the second diagram is used to explain the four relationships, and the second figure is from Patent No. 7,137,803, "Apparatus for processing a substrate comprising: a moldable layer disposed on a surface of the substrate; A lithographic method for forming a pattern in a film on a substrate, including the steps of: depositing a film on the substrate; obtaining a mold of a © stiff material which is A first patent semantic structure (21) and a second patent semantic structure (22) developed by the hard relative to the film, the mold having a first protruding feature. Wherein, each circular point in the second figure represents an inventive component, A0 (211) represents Apparatus, A1 (212) represents a substitute, A2 (213) represents a moldable layer 'A3 (214) represents m〇ld, A4 (215) Representing surface, A5 (216) represents molecular surface and A6 (217) represents pattern; while B0 (221) represents lithographic 12 200939045 method, B1 (222) represents pattern, B2 (223) represents mold, and B3 (224) represents film. And B4 (225) stands for first protruding feature. The solid arrow (218) in the second figure indicates the affiliation, for example, in the second figure, the patent 7137803, according to the description of the claim, the inventive element * A1 (substrate) is composed of the inventive element A2 (m〇ldable layer) and ** A3 (mold) constitutes 'and this affiliation is indicated by solid arrows. The dotted line (219) indicates the relationship of the connected words. For example, the invented elements A3 (mold) and A5 (molding surface) in the figure are used to "connect" the two inventive elements. If there is a verb link between the two invention elements, usually a synthetic or manufacturing method, the dotted line (226) is used to represent the relationship of the verbs. For example, there is manufacturing between the invention elements AO (Apparatus) and A1 (substrate). The relationship between the inventive element BO (lithographic method) and B1 (pattern) has a forming relationship. Finally, many aspects are often described when writing a request. These patterns can be replaced with each other. The solid line (23) indicates that the two inventive elements are equivalent. However, when the system deals with such peer-to-peer relationships, the invented elements of the peer relationship are combined and processed in a single node, indicating that the two invented elements are substantially identical and can be replaced. In constructing the semantic structure, each node is described by the vocabulary of the component of the request invention, and the vocabulary of the inventive component is usually the usage of the wise and comparative bits. In this module, the system is combined with the content of the patent specification to expand the semantics of each node to assist in subsequent analysis and comparison. Therefore, first create a Stopword list file to filter out the vocabulary that is not important in the description of the description, and then use Porter Stemming to remove the suffix of the word suffix, such as the "ing" of the progressive type or the "ed" of the past tense. . In this way, when vocabulary statistics are performed, they will not be treated as different words because of the same type. After the processing by Stopword and Stemming, the system then uses TF (Term Frequency) to calculate the word with a higher frequency in the description content and takes a threshold value as the selected standard. Finally, the vocabulary of the booklet will calculate the mutuai information along with the vocabulary in the semantic structure, as in the following formula:
= P(^)1〇g^) 在公式中’ X表示在語意架構中出現的詞彙，y表示在 說明書的詞彙；P(x)表示X出現的獨立機率，p(y)表示y出 現的獨立機率，P(X，y)表示乂和y 一起出現的機率。如果 P(x，y) = p(x)p(y)表示這兩個事件彼此是獨立。系統會給定 一門檻值，如果說明書的詞彙超過門檻值則會被列入组 架構的節點中。= P(^)1〇g^) In the formula 'X denotes the vocabulary appearing in the semantic structure, y denotes the vocabulary in the specification; P(x) denotes the independent probability of occurrence of X, and p(y) denotes the occurrence of y Independent probability, P(X,y) indicates the probability of occurrence of 乂 and y together. If P(x,y) = p(x)p(y), the two events are independent of each other. The system will give a threshold, and if the vocabulary of the specification exceeds the threshold, it will be listed in the node of the group structure.
在經過語意架構詞彙的擴充後，語意架構中每一個節 點都會有-組難來插述，而在專利語意架構的比對模組 2，針對目標語意架構跟前案語意架構做結構的相似度進 行比較。兩兩節點相似度的比較會利用 similarity，如下面公式去計算： 200939045 Σ w cos, 在公式中，ni，πμ分別表示在目標語意架構和前 ^語意架構的節點，〜和、表示在jth的權重。而詞彙 的權重會依照其出現的位置而有不同的權重如果詞 ώ現在語意架構中並為發明元件，則權重為3 ;如果 Ο ώ現是因為對等關係而出現在語意架構中，其權重為2; 如果祠彙出現在說明書内容中，則權重為i。 語意架構間在比較的時候會以目標語意架構為主，如 果兩兩節點相似度高過於門檻值，前案語意架構的節點就 會被歸到目標語意架構的節點。 如第三圖顯示語意架構的連線關係分析圖，利用相似 度公式計算後，在前案語意架構的節點編號3 (3〇1)和編 號5 (302)對應到目標語意架構的節點編號1 (303)中；前 案語意架構的節點編號2 (304)對應到目標語意架構的節 點編號4 (305)中。系統允許下面兩種情況發生，第一， 某些節點不管是目標語意架構或是前案語意架構沒有其 . 對應的節點；第二，某些節點不管是目標語意架構或是前 案δ吾意架構可對應到多數個節點。當計算完目標語意架構 與則案s吾意架構節點的對應關係後，前案語意架構就會依 照對應關係按目標語意架構的順序轉換成關係矩陣 (hierarchy matrix)。 15 200939045 如第四圖所示，因為前案語意架構節點編號3和 5對應到目標語意架構的節點編號i ，而前案語意架構節 點編號1對應到目標語意架構的節點編號2，所以前案關 係矩陣的排序從節點編號3和5 (401)開始，第二個為節 點編號2 (402)等依此類推。關係矩陣節點排序決定^， 、 訂來就要看1 吾意架構節點間是否有連線關係。關係矩陣 的連線必須考慮方向性，但可以不用是直接連線例如， 目標語意架構的節點編號1有關係連線到節點編號3，但 Q 卻不能反過來。 系統給予不同的關係連線，如從屬關係、動詞關係、 連接關係及對等關係不同的權重，亦依據連線關係的遠近 給予不同權重。節點編號丨連線到編號2是屬於直接連 線’所以權重較高為3。節點編號1連線到編號3是屬於 間接連線，所以權重遞減為2，依此類推關係越遠連線的 權重越低。建立完目標關係矩陣(target hierarchy matrix) 以及刖案關係矩陣(pri〇r art hierarchy matrix)的連線關係 ❾ 後，系統會做AND運算，如果結果不等於零，就表示兩 發明架構有相似的結構。 在檢索相似的專利前案技術模組中，系統會計算目標 - 發明架構以及前案發明架構節點以及連線的相似度來篩 *- 選相關的專利前案技術。節點的相似度用值決定，連 線的相似則用一變數c來表示，而其值為cpri()rart/c:target。 在比對排序的公式如下： Λ = (工COS nt,mk ) 16 200939045 /表示相似的語意結構。舉第四圖為例，目標發明結 構中{«2，〜}，{«2，〜}和{〜，〜}跟前案發明架構{m2, 叫}，{所2, 和{叫，讲5}相似，故兩發明結構相似度及等 於 | (C0S «2 ,,„2 + C0S «4 4 ) + | (C〇S «2,/«2 + C〇S «5，^5 ) + | (COS „4 m4 n5，m5After the expansion of the semantic structure vocabulary, each node in the semantic architecture will have a difficult group to interpret, and in the patent semantic architecture comparison module 2, the structural similarity between the target semantic structure and the former semantic structure is performed. Comparison. The similarity of the two nodes will be compared using the similarity, as shown in the following formula: 200939045 Σ w cos, in the formula, ni, πμ respectively represent the node in the target semantic structure and the former ^ semantic structure, ~ and , expressed in jth Weights. The weight of the vocabulary will have different weights according to its position. If the word is in the semantic structure and is the invention component, the weight is 3; if Ο is now due to the peer relationship, it appears in the semantic structure, its weight Is 2; if the sink appears in the content of the manual, the weight is i. The semantic architecture will be dominated by the target semantic structure. If the similarity between the two nodes is higher than the threshold, the nodes of the former semantic structure will be assigned to the nodes of the target semantic architecture. As shown in the third figure, the connection relationship analysis diagram of the semantic structure is calculated. After the similarity formula is used, the node number 3 (3〇1) and the number 5 (302) in the former semantic structure correspond to the node number 1 of the target semantic structure. (303); node number 2 (304) of the former semantic structure corresponds to node number 4 (305) of the target semantic architecture. The system allows the following two situations to occur. First, some nodes do not have a target semantic structure or a pre-sentence semantic structure. The corresponding node; second, some nodes are either the target semantic structure or the previous case. The architecture can correspond to a large number of nodes. After calculating the correspondence between the target semantic structure and the case structure, the semantic structure of the former case is converted into a relational matrix according to the order of the target semantic structure according to the correspondence. 15 200939045 As shown in the fourth figure, because the former case semantic architecture node numbers 3 and 5 correspond to the node number i of the target semantic architecture, and the former case semantic architecture node number 1 corresponds to the node number 2 of the target semantic structure, the previous case The ordering of the relationship matrix starts with node numbers 3 and 5 (401), the second is node number 2 (402), and so on. The relationship of the relationship matrix node determines ^, , and the subscription depends on whether there is a connection between the nodes of the architecture. The connection of the relation matrix must take into account the directionality, but it may not be a direct connection. For example, the node number 1 of the target semantic structure is related to the node number 3, but Q cannot be reversed. The system gives different connection lines, such as affiliation, verb relationship, connection relationship and peer-to-peer relationship, and also gives different weights according to the distance of the connection relationship. The node number 丨 is connected to the number 2 is a direct connection' so the weight is higher to 3. Node number 1 is wired to number 3 to belong to the indirect connection, so the weight is decremented to 2, and the farther away, the lower the weight of the connection. After establishing the connection relationship between the target hierarchy matrix and the pri〇r art hierarchy matrix, the system will perform an AND operation. If the result is not equal to zero, it means that the two invention architectures have similar structures. . In the search for a similar patent pre-technical module, the system calculates the target-invention architecture and the pre-invention architecture node and the similarity of the connection to screen the related patented technology. The similarity of the nodes is determined by the value, and the similarity of the connections is represented by a variable c, which is cpri()rart/c:target. The formula for sorting in the comparison is as follows: Λ = (Work COS nt, mk ) 16 200939045 / Indicates a similar semantic structure. Take the fourth figure as an example, the target invention structure {«2,~}, {«2,~} and {~,~} with the previous case invention architecture {m2, called}, {2, and {call, speak 5 } is similar, so the structural similarity and equality of the two inventions | (C0S «2 ,, „2 + C0S «4 4 ) + | (C〇S «2,/«2 + C〇S «5,^5 ) + | (COS „4 m4 n5, m5
+ C0S 〇 〇 故而，本發明在於提供一種比對專利文件的方法，用 於比對第一專利文件與第二專利文件，其中第一專利文件 與第二專利文件分別包含一組第一詞彙與一組第二詞彙。 本發明利用第一詞彙建立第一語意架構；利用第二詞彙建 立第二語意架構；比對第一詞彙與第二詞彙或第一語章架 構與第二語意架構。其中，第一詞彙與第二 件中可以定義發明元件的詞棄或是與發明元:相= ，二並，用詞彙間的語義連結_，建立語意架構，透過 第-趣與第4㈣相城計算以及第 二語意架構的相似度，輯算得出文件比對的=構與第 本發明第二較佳實施例在於可透過第 彙的相似度計算，以計算得h件比對的結果第一河 本發明第二較佳實施例在於—種八 第二構想所述之方法，兮辟六 存媒”，儲存第一及 憶體、光碟健存模組等習知之鍺存媒介碟機、快閃記 17 200939045 【圖式簡單說明】 第一圖係一系統流程圖，用以說明據本發明之一較佳 實施例之文件比對系統； 第二圖係一語意架構範例圖，用以說明語意架構表 示方式； 第三圖係一文件比對範例圖，用以說明文件比對方 式；以及 第四圖係一文件比對關係矩陣範例圖，用以說明文 件比對轉換成關係矩陣的方式。 【主要元件符號說明】 1 系統流程圖 101 第一專利 107 潛在專利前案技術（第二專利） 110 專利 108 目標語意架構 103 前案語意架構 102 步驟 104 步驟 105 步驟 106 步驟 109 步驟 Π1 步驟 Ο 18 200939045 2 語意架構的分析圖 21 第一專利語意架構 211〜217 詞彙 218 從屬關係 219 連接關係 22 第二專利語意架構 221〜225 詞彙 226 動詞關係+ C0S. Therefore, the present invention provides a method for comparing patent documents for comparing a first patent document with a second patent document, wherein the first patent document and the second patent document respectively comprise a set of first words and A set of second words. The present invention utilizes the first vocabulary to establish a first semantic structure; the second vocabulary establishes a second semantic structure; and compares the first vocabulary with the second vocabulary or the first linguistic framework and the second linguistic framework. Among them, the first vocabulary and the second vocabulary can define the word discarding of the invention component or the invention meta: phase =, second, and use the semantic link between vocabulary _ to establish a semantic structure, through the first interest and the fourth (four) phase city Calculating and similarity of the second semantic structure, calculating the comparison of the file = the second preferred embodiment of the present invention is based on the similarity calculation of the first sink, to calculate the result of the comparison of the h pieces. The second preferred embodiment of the present invention resides in the method of the second concept of the eighth, and the storage medium of the first and the memory, the optical storage module, and the like, and the flash memory 17 200939045 [Simplified description of the drawings] The first figure is a system flow chart for explaining a file comparison system according to a preferred embodiment of the present invention; the second figure is a semantic example diagram for explaining the semantic structure The third figure is a file comparison example diagram for explaining the file comparison mode; and the fourth figure is a file comparison relationship matrix example diagram for explaining the manner in which the file comparison is converted into the relationship matrix. Main component No. Description 1 System Flowchart 101 First Patent 107 Potential Patent Proceedings Technology (Second Patent) 110 Patent 108 Target semantic structure 103 Pre-sentence semantic structure 102 Step 104 Step 105 Step 106 Step 109 Step Π 1 Step Ο 18 200939045 2 Meaning Analysis of the architecture Figure 21 First patent semantic structure 211~217 Vocabulary 218 Dependency 219 Connection relationship 22 Second patent semantic structure 221~225 Vocabulary 226 Verb relationship
23 對等關係 3 意架構的連線關係分析圖 301〜305 節點 4 語意架構的關係矩陣圖 401〜402 節點23 Peer-to-peer relationship 3 Connection structure analysis diagram of 301 architecture 301~305 Node 4 Relational matrix diagram of semantic architecture 401~402 nodes
1919