TW420778B - An information retrieval system realized by fuzzy neutral network model - Google Patents

Abstract

An intelligent information retrieval system queries the matching information component according to the query words inputted. The system includes a thesaurus procedure part which searches the index key words with same meaning by the query word; an information index procedure part which searches the matching information component according to the mentioned index key word; a component sequence filtering procedure part which sorts and filters the searched information component and outputs the information component available for selection; a personalized thesaurus modification part which modifies the thesaurus fuzzy mechanism of the thesaurus procedure part according to the selected information component; and a personalized filtering modification part which modifies the sequence filtering fuzzy mechanism of the component sequence filtering procedure part according to the selected information components. The thesaurus procedure part and the information index procedure part mentioned above is consolidated by quasi-neural network model to achieve parallel fast processing and automatic learning. Besides, the thesaurus procedure part provides the query word coding and position shift compensation for input fault tolerance capability.

Description

Λ7 B7 423 ^ 78 V. Description of the invention (1) II--. H * 1-1 · 'ΙΊ I In I In I 1 — (Please read the notes on the back before filling this page) The present invention relates to a An information retrieval system, especially an information retrieval system implemented using a fuzzy neural model _ral network model, can build a personalized synonym dictionary (thesaurus), and can provide operator input fault tolerance Ability and fuzzy query, at the same time, according to the retrieval preferences of the operator, prioritize the information retrieved by the system, and provide the function of quick selection. Printed by the Consumer Cooperatives of the Central Bureau of Standards of the Ministry of Economic Affairs. Due to the development of the Internet and information, the acquisition of information has become very diverse. For an operator, how to effectively find out the parts of his own interest from a lot of information is a very important issue. For example, for software programmers who use object-oriented (〇bject_〇riented) technology for software development, when they receive a demand for a large-scale software development, they most want to go through a requirement analysis process. 'After obtaining the formal specifications, you can directly access the appropriate, developed software components to quickly combine them into a software prototype' for testing. As for other situations such as public library query systems, WWW (World Wide Web) search tools, and other information services (such as a large number of news, management and query of legal documents), there are also similar requirements, and operators most want It is how to effectively capture the information needed to meet the needs of their own work. However, when the amount of information increases rapidly, or the number of software components contained in the software component database (ibraries) increases, the entire process of handling information and software components is full of paper. Guanjia County (⑽) Α4 · (210χ297 公 幻 420 * 78 Λ7 ~ ——-~ __ V. Description of the Invention (2) ''-Certainty. To achieve a better recording effect, it often depends on the operation The user's own test ㉟, that is, whether the query word entered by the operator can accurately describe the characteristics of the required information. In view of this, the 'reliance acquisition system' has become an important subject in information management and information reproduction technology. In the traditional information acquisition system, it is usually established based on the structure of Boolean logic. That is, the query word entered by the operator is used to compare with the characteristic key value (keywo) of the information or software components to determine that it meets the needs of the operator. However, this traditional information system is in use. The flexibility is very poor, and its disadvantages are described as follows: l. In a typical information age system, the system usually hopes that the query entered by the operator when querying the database can fully match the keyword officially provided by the system. These formal _ words are set by the system when it is set up to describe the characteristics of the document or software component to be retrieved. However, in practice, it requires the operator to be able to use the keywords stored in the system accurately. According to this shortcoming, the solution proposed by the conventional technology is to build a synonym dictionary (thesaurus) for the keywords of the system. That is, each keyword in the system can be in the synonym A group of synonyms is defined in the dictionary. As long as the operator enters one of this group of synonyms, it can be inferred that the input is the corresponding _ word. This -square cut adds The operator can enter the vocabulary of the query 'but it does not actually solve the problem. Because this type of synonym dictionary is generally used in a fixed way, the table lists the keywords and the 5 paper standards are applicable to the Chinese national standard (CNS ) Α4 size (210 × 297 male thin) (Please read the precautions on the back before filling in this page) J-^-I ------- II --------.! Central Ministry of Economic Affairs Standards Bureau, Masonry Consumer Cooperatives, India 9 9

ΑΊ Β7 5. 'Invention description (3 relations'), so it is not possible to adjust the synonym dictionary relationship for the operator's search. ~ Phase 2 Β #The system's f-message operation system is based on the query 7 entered by the operator. It ’s true that there is no tolerance for erroneous input in the conventional system. It is difficult to guarantee that when a person enters an input action, they can completely lose the input by mistake. In addition, it may also be a synonymous but different word. For example, 'the operator may have missed the first-mother in the input of a query word 2; or the synonym dictionary originally used a certain-word verb, but the operator lost Its noun. The situation like this member 'may make the system unable to find the corresponding keywords correctly. According to this-the shortcomings, the conventional technology is generally compensated by using the substitution table (e). The substitution table lists the possible error forms of the key words in each synonym dictionary ’and possible part-of-speech changes, so as to correct errors that may be caused when the operator turns. However, this kind of substitution is inherently unsuccessful, and it is basically impossible to handle all possible errors. Du Yinju, Employee Cooperative Cooperation of the Central Bureau of Standards, Ministry of Economic Affairs · The conventional information retrieval "Unable to determine the information to be processed", that is, the fuzzy query processing cannot be provided. In other words, the operator must accurately describe the query information Documents or software components can only find the required data. In other words, the operator must have a considerable degree of understanding of the information file or software component to be queried. 'Otherwise, the wrong information often leads to the conclusion of the query. No original expectations. W Standard (⑽} 2 " 97 公 爱) A7 423778 V. Description of the invention (4) 4 · The last thing to mention is that 'traditional information retrieval system based on certain specific keywords' is usually A considerable number of matching documents or software components can be found. Generally speaking, 'more advanced information retrieval systems can usually rank and output the found documents or software components based on their relevance to query words. However, This sorting and filtering function is quite fixed, that is, the same sorting is used by any operator And filtering procedures, so it is quite inflexible in use. According to the above, it can be seen that although the existing information retrieval system can basically achieve the purpose of query and retrieval, it is not very convenient in use. It is difficult to find a suitable document or software component. This is also the problem that the present invention intends to solve. In view of this, the 'main purpose of the present invention' is to propose an intelligent information retrieval system, which has a personal preference for the operator. A synonym dictionary is set, so it is possible to flexibly perform a personal synonym conversion based on a query keyword 'input by an operator and then perform a query. Another object of the present invention is to provide an intelligent information retrieval system , Can provide the ability to input fault tolerance, that is, even if the operator made a certain degree of error in the input, the system can compensate and query. Another object of the present invention is to provide a smart information operation Acquisition system, which can provide the performance and processing of uncertain information, that is, query in a fuzzy manner, Plus the flexibility in use. Another aspect of the present invention is to provide an intelligent system for picking up information, which can apply Chinese national standards (CNS) to the paper size retrieved by the system according to the retrieval preferences of the operator. ) A4 size (210X297mm) 'n · I--n I n In-i 1-' -I 1. Ϊ́ i I I--I. IIII 1 (Please read the precautions on the back before filling this page) Printed by the Industrial Standards and Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs, and printed by the Central Samples and Standards Bureau of the Ministry of Economic Affairs, the Consumer Cooperatives of the Central Standards Bureau. Increase the convenience and function of selection. According to the above purpose, the present invention proposes a smart information building system for querying the corresponding information element according to the input query word. 2. The g subroutine, which uses the query word fuzzy correspondence mechanism, uses this query word to search for at least the _ index key to the synonymous relationship. | The program department is used to search for at least one person corresponding to the search key η 4 ^ 7C, a software ordering filtering process 4 'It is based on the-fuzzy ordering mechanism, will search out The information elements are sorted and filtered to output the desired information elements for selection; a humanized synonym adjustment unit that adjusts the synonym word correspondence mechanism of the above synonym program procedures based on the selected f-sign elements; and A humanized filtering adjustment unit adjusts the fuzzy sorting and data collection mechanism of the above-mentioned component sorting and program part according to the selected information components. In the embodiment of the present invention, a software component that has been packaged to perform a specific task independently is used as the f-signal component described above. In addition, the above-mentioned synonym program section and information index program section are composed of neural-like networks to achieve parallel and fast processing. In addition, the synonym word unit uses the query word encoding and position offset, which can be called the input fault tolerance capability. The synonym program unit described above can operate in two modes, which are recaU phase and learning f mode (phase). In the recall mode, 'this synonym program department uses a coding department to compile this query, and generates a query word encoding style, which corresponds to this query (nn | _1 .m —JJ— II ί I --- I — .1 Ding, -e (Please read the notes on the back before filling out this page) 5. Description of the invention (6) The characteristics of the query word. Then use a similarity calculation section to use this query word coding style and The system synonym encoding styles set by several systems are compared to generate corresponding similarity values to indicate the similarity between the query word and these system synonyms. These system synonyms correspond to each other Keywords set in several systems. Based on the above-mentioned similarity value 'and the correlation value between the system synonym and the system keywords, a correlation section is used to compare with a predetermined threshold value, so as to derive from these system keywords In the selection, the output is selected as the above-mentioned index key. In practice, the 'similarity calculation section may be composed of nodes of a neural network, and each node corresponds to a synonym set by the system, and is associated with It can also be composed of nodes, corresponding to keywords set by a system. Printed by the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs (please read the precautions on the back before filling this page) In the learning mode, the synonym program department Using the encoding unit, the query word and the extracted component keywords corresponding to the selected information component are respectively used to generate a query word encoding style and an extraction component keyword encoding style. The similarity calculation unit respectively uses this query word The encoding style is compared with the system's synonym word encoding style to generate the corresponding group of similarity values and the second group of similarity values. The association unit is based on the first group of similarities Degree value and the second set of similarity values, and the corresponding output values of the system word and the system_word-like original correlation value 'are calculated. Finally, the reuse-judgment section, according to the first, and the similarity value , The second group of similarities and the output value, to determine whether to modify the similarity calculation unit and the associated unit. The information indexing program includes a Keyword temporary storage department, in order of 9 paper rule Lai Cai Guan Jia Su 7 ^) weaving grid (21 () > < 297 public -------- 4207T8 Λ7 _____B7 V. Description of the invention (7 ) ^ 'Store index keys; and a plurality of component index units, each corresponding to each information element, and storing a component key corresponding to these information elements' and a corresponding information element identification code. When an index key is received' It is used to calculate the similarity between the stored component keywords and determine whether it exceeds a predetermined threshold to determine whether to output the corresponding information component identification code. Brief description of the drawing: In order to make the above-mentioned object of the present invention, Features and advantages can be more obvious and easy to understand 'The following is a detailed description of a preferred embodiment in conjunction with the accompanying drawings: Figure 1 shows a block diagram of the functional structure of the intelligent information retrieval system in the present invention. FIG. 2 shows a schematic circle of an example of a query screen of an information retrieval system in this embodiment. FIG. 3 is a block diagram showing the functional structure of the synonym program unit and the personalized synonym adjustment unit according to the embodiment of the present invention. Printed by the Consumers' Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs (please read the precautions on the back before filling this page). Figure 4 shows a schematic diagram of an example of the coding procedure in this embodiment. FIG. 5 shows a detailed structure diagram of a synonym program part using the structure of a fuzzy neural network according to an embodiment of the present invention. Fig. 6 is a block diagram showing a functional structure of an information indexing program unit according to an embodiment of the present invention. Fig. 7 shows a detailed structure diagram of the information indexing program part using the structure of the fuzzy network via the embodiment of the present invention. Explanation of symbols: This paper size is applicable to China National Standards (CNS) A4 (210x29 ·? Gongchu) 420778 Λ7 Instruction B7 of the Consumers' Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs 5. Description of the invention (8) ~~~~~~ 1 ~ synonym program section; 3 ~ information index program section; 5 ~ component sorting filter program section; 7 ~ personalized sort filter adjustment section; 9 ~ personalized synonym adjustment section; 10 ~ operator; 11 ~ query word; 13 ~ index key: 15 ~ software component identification code; 17 ~ sorted filtered software component identification code; 19 ~ retrieved component keyword; 21 ~ retrieved software component; 62 ~ software component function query item; 64 ~ Related Media Query Items; & ~ Software Component Type Query Items; 68 ~ System Type Query Items; 101 ~ Compilation Department; 103 ~ Similarity Calculation Department; 105 ~ Correlation Department; 107 ~ Judgment Department; u, L2, L3, L4, L5, L6, L7 to processing layers; Lla, Lib, L61-L6m to sublayers; 301 to index key temporary storage unit; 303 to software component indexing unit. 'Embodiment: This embodiment Disclosed information retrieval system, using software components as pages Take the elephant system as an example. The so-called software component here refers to the package of the source code and the private data used for the program function to be packaged into an independent and complete certain function. Components, such as software components used in object-oriented technology. However, for those who are familiar with this technology, they can use the same technology disclosed in this embodiment to expand their use in capturing various documents, patterns, animations, videos. In the case of multimedia data such as songs, songs, etc., the application is not limited to the environment described in this embodiment. According to the above description of the conventional technology, the information in this embodiment (referred to as the software component in this embodiment) retrieval system To be able to achieve the following purposes: First, according to the operator's usage habits and personal preferences, flexibility 11 This paper size applies Chinese National Standard (CNS) A4 specifications (210X297 mm> 1----= -1--1 ---1 clothing ----- -I In i (Please read the notes on the back before filling out this page) Printed by the Consumers' Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs 420778 at _____ Explanation (9) to adjust the synonym dictionary to facilitate the query; second, it can provide the ability to input fault tolerance; third, provide the performance and treatment of uncertain information, that is, provide a fuzzy query method; fourth, the basis Operator's retrieval preferences and usage habits 'to sort the extracted software components and filter the less relevant software components to increase the operator's convenience in selection "In the present invention,' fuzzy information extraction is used Techniques such as fuzzy information retrieval, knowledge engineering, machine learning, etc., achieve the above purpose. The embodiments of the present invention will be described in detail below with reference to the drawings. Fig. 1 shows a functional block diagram of an intelligent software component retrieval system. As shown in Figure 1, the entire software component extraction system is composed of a synonymous subroutine section 1, an information indexing program section 3, a component sorting transition program section 5, a personalized sorting transition adjusting section 7, and a personalized synonymous adjusting section 9. . The function can be divided into two parts, that is, the query retrieval mechanism composed of the synonym program section 1, the information index program section 3, and the component sorting filter program section 5 in the upper half of the figure, and the query retrieval mechanism in the lower half. The adjustment mechanism formed by the personalized sort filter adjustment unit 7 and the personalized synonym adjustment unit 9. In the process of query retrieval, the operator 10 first issues the query word 11 'and the synonym program unit 1 is responsible for resolving the corresponding keywords. These keywords are set by the system and are used to describe the characteristics of the software components that can be manipulated. In this embodiment, the query word 11 issued by the operator 10 includes four query terms (facets), which are used to define the software components to be retrieved by the operator. The four inquiries are: Software component function. This paper size is applicable to China National Standards (CNS) A4 (210x297 mm) (Please read the precautions on the back before filling this page.) Order the staff of the Central Bureau of Standards of the Ministry of Economic Affairs. Printed by the cooperative k. V. Description of the invention (10) (function), software component type (file type), associated medium (medium) and system type (operator can set these four query items to define the required Software components. The query term of the software component function is used to define the function and function of the software component. In this embodiment, the query term of the software component function can be entered by multiple query words' and then distinguished by a separator. The software component type query is used to define the type of a software component, that is, what type of software component, such as 〇CX, VBX (visual basic) ^ VCL ^ DLL (dynamic link library) 'C ++, delphi, etc. 》 The related media query term is used to define the device that a software component acts on, such as disk, keyboard, port 'neUvrk, screen Interface card, etc. System-type queries are used to define the operating platform for a software component, such as

Windows 3.x / 95 / NT, DOS, OS / 2, UNIX, and more. In this embodiment, the query items that have only software component functions are input by the operator themselves, and other query items are provided by the system with a fixed keyword list for the operator to choose. Fig. 2 is a diagram showing an example of a query screen of an information extraction system in this embodiment. As shown in the figure, 'Operator 10 can use the settings of each query item to define the required software components, where 62 represents the software component function (64) and the associated medium 丨 then immortal !!!), which represents the software Component type (file type), 68 indicates system type. It can also be clearly seen from the figure that the operator can freely enter the query word used to define the function of the software component, and the other query terms are directly applicable to the national standard (CNS> A4 specification (210X297)) of this paper size ( Please read the notes on the back before filling in this page} Λ7 42D778 V. Description of the invention (ii) It is selected from the keywords set by the system. Therefore, in this embodiment, synonym processing is performed for the query words entered by the operator. 'Fault-tolerant processing, etc.' mainly refers to the query word entered in this query item. After query word 11 is entered into the synonym program unit 1, the analysis of synonym words is performed, corresponding to several keywords that were originally defined by the system. In this example, it is called an index key (indexe £ j key word). 13. An index key is a keyword originally defined by the system, which has a certain synonymous relationship with the input query word. The synonymous relationship is defined by the synonymous program unit 1. In this embodiment, the synonymous program unit 1 is composed of a fuzzy neural network, and the query word encoding (encoding ) And position offset compensation methods to provide the ability to input fault tolerance. As for the detailed structure of the synonym program unit 1 will be described in detail later. In addition, as described above, the synonym program unit in this embodiment is only for software The query words of the functional query items are processed synonymously. As for the query words entered by other query items, they are directly selected from the keywords preset by the system, so there will be no problems with synonyms and input errors. According to the synonyms program 1 The searched index key 13 is used by the information indexing program section 3 to retrieve the corresponding software components. In this embodiment, the 'information indexing program section 3 will identify the software component's identification number 15 In addition, the information indexing program section 3 of this embodiment is also composed of a fuzzy neural network, so the efficiency of parallel processing can also be improved. The details will be detailed later. Finally, the information indexing program section 3 outputs Identification of software components is exhausted. This paper size is applicable to China National Standard (CNS) A4 specifications. (Please read the precautions on the back before filling This page} · The consumer cooperation of the Central Bureau of Standards of the Ministry of Economic Affairs, printed by DU 420 420778 V. Description of the invention (12) One *-15 is sent to the component sorting program section 5, sorted ⑽㈣) and After filtering, the output is selected on the display device. General query words can be found in a considerable number of software components after processing by the synonym program unit i and the information index program unit 3 described above. Query terms, and as the number and scope of software components that can be retrieved increase, the number of matches will increase. It is inappropriate for the operator to freely find the required software components from so many messages, so It is necessary to sort out and filter out the soft components. The purpose of the sorting process is to require all extracted software components to be displayed in order for the operator to choose according to the relevance to the query word and the operator's past choice of software components, thereby allowing the operator to find The required plutonium element. The purpose of material processing is to reduce the number of software components with low relevance, so that the operator is more tolerant and finds the required software components. Therefore, a critical value 可由, which can be adjusted by the operator, is set in the component ordering program section 5 so as to transition software components with low relevance. The operator 10 can select the actual software component 21 according to the displayed software. Another feature of the software component extraction system printed by the Central Bureau of Standards, the Ministry of Economic Affairs, Shellfish Consumer Cooperative, is that it has an automatic feedback adjustment mechanism, which can choose the actual needs and software according to the operator 10; The retrieved component 21 and its keyword 19, the displayed operating habits and retrieval preferences, respectively, adjust the component sorting and filtering program section 5 and the synonym program section 丨. The personalized sorting and filtering adjustment section 7 adjusts the sorting mechanism and filtering mechanism in the component sorting and filtering program section 5 according to the capturing component 21 so that it has a learning function to match the actual work. Paper ruler 15 (210X297 mm) Λ7 B7 423778 V. Description of Invention (13) Need. Personalized synonym adjustments. For free 0ρ9, according to the software component keyword 19, adjust the synonymous relationship in the synonym program 1 to make it have a learning function to meet the needs of actual work. I-I n- In Mil ------ Order (please read the notes on the back before filling this page) The following diagrams are used for the synonym program section, the information index program section, and the component sorting and filtering program. Section 5 explains.

Jj synonym program section (Π Figure 3 shows the functional structure block diagram of the synonym program section 1 and personalized synonym adjustment section 9 of this embodiment. As shown in the figure, the synonym program section! Includes the encoding section 101, similar Degree (similarity) calculation unit ι03, association unit 105, and judgment unit 107. The synonym program unit 1 can be divided into two modes (Phases) in operation, which are the recall mode (recaU and the learning mode ( learning phase). Recall p'hase: In the recall mode, the Central Consumers ’Bureau of Standards and Labor ’s Consumer Cooperatives prints that in the recall mode, the query word u entered by the operator is first encoded by the encoding unit 101. In this implementation In the example, the encoding format mainly represents the number of letters and the first occurrence position of the word to be encoded (the query word at this time) in a parallel manner. The following is the encoding method used in this embodiment: (a)- At the beginning, all the letters of the word to be coded are converted to lowercase, and some non-a to z and blank words are converted to blank words. (B) Set up 27 registers, corresponding to the English letters α to z and blank words. c) After calculating the conversion 'Number of each letter in the word to be coded, and 16 paper sizes are applicable to the Chinese National Standard (CNS) A4 specification (21ΰχ 297 mm) 42JT78 Λ7 V. Description of the invention (14) — ~ and store the amount of these letters in $ To the corresponding register. (D) At this time, there should be many registers that have not yet been filled with data. Therefore, these unfilled registers are used to store each letter in the word to be encoded after the change. (Including blank words) the first occurrence position. The stored = formula is to calculate the relative position of each letter in the word to be encoded, and it becomes a negative value stored in the adjacent temporary register stored in the corresponding letter register. When saving, it is preferentially stored in the left register. If the left register has been filled with data, then choose to save it in the right register. If both registers are filled with data, this position value is discarded. , ⑷Even if the location data is filled in, these 27 registers may still have unregistered data registers. At this time, all the letters of the word to be encoded after the replacement (that is, only & to z and blank) "), Fill in all the temporary At this time, the encoding action of the word to be encoded is completed. Figure 4 shows the intention of encoding "邛 _c_r〇i" in this implementation to perform encoding. The fourth circle ⑷, ⑻, ⑷, and ⑷ Respectively indicate the content value of each register in the above steps ⑻, =, ⑷, and ⑷. Among them, the data, 〇 ”means not: the state of the real body 1 26 registers corresponding to each other Matters needing attention are re-filled (" this page) The Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs are printed on Ί ",, 'e"',, Ί, "k", "〇, ',,, v" ,, n , "Y '77x," "corpse, 7a,: ,, s, z, b'

'T w, g ”” t' ”c, MU,“ 11 ”,)”, 4, ’, and 5, and the 27th register corresponds to a blank word. This letter position assignment :::: Code characteristics' and the most commonly used letters are arranged L = ::: next to the rule, plus the recorded information records 17 paper ruler Cai Caiguan Λ7 420778 V. Description of the invention (15) ~ --- -The query word encoding format generated by the encoding unit 101 will be sent to the similarity calculation unit 1G3 to calculate the similarity with the synonymous word of the line shirt; then it will be sent to the association unit 1G5 to determine the actual and query words What are the relevant system keywords, and the output is used as the index key? 13 〇 In the similarity calculation unit 103, the encoding format of the synonym defined by the system is stored in advance. Its encoding method is the same as the above. Synonyms correspond to the “keywords” defined by a certain system—the keywords are used to directly describe the functions of software components (in the case of this embodiment). Therefore, in the similarity calculation unit 103, the query word u and the system are first used. Compare synonyms to produce two The similarity value between the two represents the degree of similarity between the two. The association unit 105 stores the synonymous word defined by the system and the system. The correlation value between the defined keywords. For example, if the system defines a certain The key word for the function of each software component is, “ab〇ut dial〇gb〇x”, which is used to describe the function of “related to the dialog box”. Therefore, the synonym may be printed as “' 'interface', 'dialog', 'help box', etc. According to experience, it can be seen that the correlation between the above two synonyms and the actual keywords is not the same. For example, `` dialog '' and `` about The correlation degree of "dialog box" should be very high, and it can be set to 0.7; but synonyms such as 'hdp b〇x,' may be less relevant, so set it to 〇5: and like "interface" Synonyms, because the meaning is too broad, may have a lower degree of correlation, so set it to a lower value of 0.3. Therefore, when the similarity value calculated by the similarity calculation unit 103 is sent to this paper scale, the country of China is applicable. Standard (CNS) A4 specification (210x297) 2077 Printed by the Consumers' Cooperatives of the Central Bureau of Standards of the Ministry of Economic Affairs B7 V. Invention description (16) After the associating unit 105, you can "Relevance value" determines the degree of dependence between the query word and each keyword. In this embodiment, "is the minimum value of the corresponding similarity value and relevance value, then determines the relationship between the query word and each keyword And then compare it with a predetermined threshold to determine whether to output this keyword as the index keyword 13. In the recall mode, the 'synonym program unit 1' does not use the judgment unit 107, nor The personalized synonym adjustment unit 9 will be used for adjustment. In addition, the synonym program unit 1 also includes a position offset calculation unit (not shown in Figure 3, whose function is implemented by the L2 layer node in Figure 5), which can be based on the encoding format of query word 11, Provides input fault tolerance. This section will be detailed later. Learning mode: In the learning mode, the 'synonym program section 1 will use the function provided by the personalized synonym adjustment section 9 to modify the similarity value calculation parameters (described later) in the similarity calculation section 103 and The correlation value in the correlation unit 105. The correction judgment is determined based on the software component selected by the operator and the corresponding retrieval component keyword 19. First, the query word 11 and the extracted component keyword 19 are input into the encoding section 101 and encoded to obtain the corresponding query word encoding format and the extracted component keyword encoding format. The query word encoding format and the retrieval element keyword encoding format are sent to the similarity calculation unit 1 at the same time, and the similarity between the two and each system's synonym is calculated separately. At this time, two sets of similarity values are generated, which correspond to the query word 1 1 and the paste element ^^^ 1 1 IH ^ I n ^ — n a ^ -5 (read the first; read the meaning of the matter and then Fill out this page}

420778 Λ7 Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs, India and India B7 V. Description of the invention (17) Keyword 19. These two sets of similarity values will be sent to the judgment unit 107 in addition to the correlation unit ι05. However, at this time, the association unit 105 performs an operation different from that in the recall mode, and then generates a set of output values to the judgment unit 107. The judging unit 107 then decides whether or not to perform the correction and how to perform the correction based on the two sets of similarity values and the output values, and the personalized synonym adjustment unit 9 is actually responsible for the adjustment operation. As for the encoding method of the encoding section 101 and the similarity calculation section 〇03, the similarity calculation method is the same as the recall mode described above. FIG. 5 is a detailed schematic diagram of the implementation of the synonym program unit 1 by using the structure of a fuzzy neural network in this embodiment. The advantage of using neural-like networks is their parallel processing and learning capabilities. As shown in Figure 5, this fuzzy neural network has five layers of nodes, which are denoted as LI, L2, L3, L4, and L5, respectively. The functions of the nodes in each layer are explained below. L1 layer In the L1 layer node ', it includes two sub-layers, namely Lla and Llb, each of which has n nodes. Therefore, nodes qi, ^ ..... qn are contained in the sub-layer Lla, and nodes ki, k2 ..... kn are contained in the sub-layer Lib. The sub-layers Lla and Lib are used to store the encoding format of the query word u and the retrieval element key 19, respectively. Therefore, n in this embodiment is set to 27, and the functions b of the knowledge points qi, q2, ..., qn and the nodes h'k2, ..., kn are similar to the temporary registers in the encoding process described above, and respectively store the query word 11 And the total number of each letter of the retrieval component keyword 19, the first occurrence position 20 The paper size is applicable to the Chinese National Standard (CNS) M specification (210X297 public trend) — ^ ν ί -I m. —II ^^^ 1 I. '·, 1 · (Please read the notes on the back before filling this page) /! R > / ·' Q i

M B7 V. Description of the invention (18) and actual letter information. In addition to the actual responsible for the module not shown in the figure, the L1 layer node is stored in the coding result of the coding section 101 in the third figure. In addition, in the recall mode, only the data of each node in the sub-layer Lla will be used, and in the learning mode, the data of each node in the sub-layer Lla and the sub-layer Lib will be used. L2 layer In the L2 layer, nodes I, p2, p3, p4, and p5 are included.

Pd 'is a synonym that is defined by the system. Therefore, the encoding format of the synonym of this system is also stored. Each node receives the encoding format of query word 11 and retrieval component keyword 19 from the L1 layer. In the recall mode order, it calculates the encoding format of query word 11 and the encoding format of the system synonym it stores. In the learning mode, the encoding format of the query word 11 and the retrieval component keyword 19 are calculated respectively, and the similarity value between the encoding format of the query word 11 and the system synonym encoding format is calculated. This similarity value represents the degree of similarity between the input word (query word U or extraction component keyword 19) and the two synonymous words. In addition, using the position offset calculated by the 12-layer node, it is also possible to find the input error part in order to compensate the input query word 11 and its encoding format. In this embodiment, each node of the L2 layer performs the following calculations: = Σ Ψα} / ^}} ⑴Ca /) = 1C Σ K) / 2} (3) The formula 表示 is expressed in a certain L2 layer On node ⑹, its storage is the same as (2) —-- I 1 I---II--I — — ί I.. _ _ (Please read the precautions on the back before filling this page) 21 papers The scale applies to the Chinese family standard (CNS) of 4 specifications (2) 0297 mm. 5. Description of the invention (19), the position offset% between the sub (Pi) and the input query u (q). ,) Where q ′ and PM are the coding resources for the query word coding format q and the peer coding format Pi, respectively, and j is Bu 27. (. It means the weight between the j-th node (屮) in the Lu sub-layer and the i-th node in the L2 layer (weight), which can also be regarded as the 7th code in the query word u The importance of the data (㈣⑽_e) ... In general, this importance can be expressed by the probability of a certain letter appearing in the general query word, for example, z is less common in the query word, so it should be more important if it appears. Therefore, the weight relative to "z" must be larger. In addition, it must be noted that formula (1) is only calculated for the negative value 'in the encoding format, that is, it is calculated only when 屮 and Pij are both negative. So, in ⑴, if 屮 is negative, ^ is equal to 1 otherwise it is 0, if

Pij is a negative value, pij 'is equal to 1 otherwise it is 0, if qj and Pu are both negative,% is equal to q "otherwise 4 0. When the position offset s (q, Pi) is greater than or equal to 1 or less than or equal to] Shaw, it means that all negative position information in the query word encoding format q must be compensated (reduced or added respectively) This offset value ... Printed by the Consumers' Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs Using the coding format and the position offset s (q’Pi) ′ calculated by Equation (0), it is possible to check whether the input query word is missed. For example, if the operator wants to input, "fuzzy", the query word, but the first letter is missed, and it is typed "uzzy". If the position offset is calculated, it will be found to be greater than or equal to 丨, which indicates that there are missing letters in the front, so the offset is compensated to the position of the encoded style, so the missing words in the query input can be solved. The case of the preceding letters. Except that 22 papers are printed from the applicable Chinese national standard (c ^ y ^^ 10x297) ^ 420Π8 Printed by the Shellfish Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs 5. Description of the invention (20) In addition, the fuzzy mechanism and encoding of this embodiment are used In order to achieve the effect of reducing the wrong input. 1 can also be used (2) is used to represent the query word (5 and synonym * / p | 3 "average Hamming distance (averageHammingdistance). Among them, the right says h ", And q" need to be explained. Prior to this, the three sets are first defined. 8. It is a set that records the number of each letter of a word in the coding style. This value is a set of less than 27 positive integers;% Is the set of negative values of the first occurrence of each letter of a _ character in the coding style, which is a set of negative integer values greater than ~ 27; Se is the set of individual letters of a word in the coding style order, which is The ASCII value is greater than or equal to the set of 32 letters. Then, there are 4 possible values of the mountain: the value of one, the absolute value of 屮, the absolute value of three, and the absolute value of the last. The following formula shows: if ⑷ e \ and (A = 〇〇r 巧 e)) or (qj = 0and (i * eSporPv eSc)) 〇r i9j e Sc and (Py = 0 or eSp)) or (9 / e Sp and PtJ e Sp and ^ PtJ) or (gj e Sc and PtJ e SQ and qy Φ Ptj ) \ q, I if (qj e S „and (Pi) e Sp or Pt) e Sc)) \ pi, I if ^ S„ and (qj e Sp or qs e Sc))-/ ^ 1 otherwise Pij Three possible values are shown below: P, wp, sn 1 if (P ^ SporPveSc)

I 1—I i n n'i —J. H-I __ I {Please read the notes on the back before filling in this page J Order d „

K 0 otherwise and q / also have three possible values, as shown in the following formula: 23 This paper size applies to China National Standards (CNS) A4 specifications (210 × 2π mm) 42Q778 A7 B7 V. Description of the invention (21) 1 li {q ^ SP0r ^ j ^ Sc) 0 otherwise Equation (3) calculates the similarity between the two based on the average Hamming distance, and + Fe respectively represents two fuzzy sub-suppliers. (Fuzzifier), which defines the range covered by the membership function of the average Hamming distance after encoding and between styles, and the sensitivity to the average Hamming distance. In this way, the similarity value of a certain query word corresponding to a certain synonym can be generated. The above describes the case where the query word 11 calculates the similarity value, but it can also be used in the case of manipulating the key of the component. In addition, the function of the L2 layer node corresponds to the similarity calculation unit 103 in FIG. 3. L3 layer In the L3 layer, a node SA is included. The node SA is used to find out the similarity value sent by each node in the l2 layer, which is most similar to the input query word u. In this embodiment, the node SA mainly plays a role in the learning mode, and its behavior is not as follows: μ (Κ, P): 气 邮 乃 ⑷ where K represents a query word 11 or a component keyword 19 d③ represents Aggregation function, in this example, the maximum function. And μ (κ, P *) indicates the similarity output of the L 2 point corresponding to the synonym I found closest to κ. In addition, in this embodiment, the SA0 outputted by the point SA is a vector. In addition to the above-mentioned pair of values most similar to the query word 11 or the manipulation element keyword 19, it also includes 24 paper scales.囷 National Standards (CNS) Specifications (210X297 mm) (Please read the notes on the back before filling this page)

, 1T 420778 A7 B7 V. Description of the invention (22) and query word 11 or extract component keywords! 9 is most similar—for L2 node w. Its function will be described later with the L5 layer, and its function corresponds to the judgment section 107 of FIG. 3. In the L4 layer, node scales, r2, R3, r4, ..., R are included.

C corresponds to the keywords (keyword) set in the system. As mentioned above, a system keyword may correspond to several synonymous words set by the system, but the correlation between each synonym and the system keyword may be different. The actions of the nodes in the L4 layer are determined according to the similarity value calculated by the above L2 layer (corresponding to a system synonym respectively), and corresponding to each L2 layer node = relevance value. Dependencies between key children. In this embodiment, the dependency relationship between the query word and the keyword can be determined by taking the smallest of the similarity values (that is, the most similar synonyms) and the corresponding correlation degree. Next, each L4 layer node will set a critical value (indicating the index of the L4 node) 'whether the system key words used in the rhyme word U_L4 section meet the requirements. If this condition is met, the L4 node can output the corresponding keyword as the cable bow I keyword 13 and send it to the information indexing program section 3 »Each L4 node has different computing characteristics in the recall mode and the learning mode . The one-time operation is performed in the recall mode, and the second min operation is performed in the sub-I mode. The operation performed by each node can be expressed by the following formula. This paper size applies to China -m--Inhr IIK ^. Ml I in m s- I (Please read the precautions on the back before filling this page) Printed by the Bayer Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs ¾. J778 A7 V. Description of the Invention (23) ~~~ '—- [Null ^ othenvise (6) ° l (^) = min (min ^, ^)) ⑺ ----------- 篆 ----- -? ^ -tl (Please read the notes on the back before filling out this page) Among them, formula (5) and formula ⑹ are executed in the recall mode and will be executed in the learning mode. In the formula (1), the 'free' means the value of the degree of synonymy of the synonymous words related to the ^ th node. & refers to the first Η f node. And L respectively represent synonymous words ^ associated with this L4 layer node, that is, L2 layer nodes ^ and & connected to the i-th L4; f node. q is the query word 11 encoding style, and k is the extraction component key M encoding style. Using Equation (5) ', the degree of dependence between the query word and the related keywords can be calculated. In the formula (6), the LR code represents an identification code index of a system keyword, and is used to refer to a real word, not its coded form. In addition, with the setting of the threshold value Θ ?, the operator can be set to use generalized servants to deal with terms, narrower terms, or normal terms. That is, when the threshold value is smaller, it means that it is processed in a broad sense; when the threshold value Θ / is larger, it means that it is processed in a narrow sense. In the formula (?), 〇 (P!) And 0 (P2) are the values output by the layer nodes related to this L4 layer node, respectively. The final print of the employees ’cooperatives of the Bureau of Decision and Standards of the Ministry of Economic Affairs, 〇R (Ri) and Ol ^) is the output of this L4 node in the recall mode and the learning mode, respectively, and sent to the information index program department 3 and L5 Layer. The L4 layer is the associated part 105 corresponding to FIG. 3. L5 layer L5 layer contains only one node RA, and it receives 26 paper sizes in the learning mode. It is applicable to the Chinese National Winter Standard (CNS) Λ4 specification (2 丨 0χ29? Gonglu). System Λ7 V. Description of the invention (24) a '-input from each node of the L4 layer. The role of the node RA is to select the largest one in the correlation (dependency program) output by the layer points, and its action can be described as follows: 0 '(5) The L5 layer and the L3 layer are in the learning mode in this embodiment, Determine whether and how to make corrections. As mentioned before, in this embodiment, the ^ 3 layer will output at least two maximum similarity values, which respectively correspond to the query word 11 (whose coding style is q) and the retrieval component keyword 19 (whose coding style is k). The L5 layer will output a maximum output value. Using these three values and two independently set thresholds 1 and 1, the adjustment operation of the personalized synonym adjustment unit 109 can be determined. (Case 1) When the output value of node RA (L5) is greater than the threshold value, or the two similarity values output by node SA (L3) are both greater than the threshold value, the following learning procedure is performed: 1 (a) For the nodes in the L2 layer corresponding to the two maximum similarity values (set to 沢〗 and W2), the stored system synonym word coding styles pW1 and pW2 are modified. The correction method is shown below, where π and π respectively represent the number of times of being changed. / f = W (furnace 1 + 1) < + 1 / (, 丨 +1) ·% pr1 = Nv2 / (JVW2 +1) · +1 / (ATW2 +1). 'ki

Nvl = jVH " + l 1 (9)

Nw2 = Nwi + l where i = l ··· n. (b) If the output value of the node RA (L5) is less than the critical value θ3, a new L4 node w3 is established, and the correlation degree value and the number of times it has been changed are stored in this paper. 2〗 (Ox297). {Please read the precautions on the back before filling this page): Clothing · 420778 V. The initial value of the description of the invention (25) is set as: Λ7 B7

W w3 = 1.0 Printed by the Consumer Cooperatives of the Central Bureau of Standards of the Ministry of Economic Affairs (13)

Nwi = \ (10) Otherwise, for the [4th layer node in the L4 layer that outputs the maximum value with min operation, the stored correlation value r3 is modified for forward learning. The correction method is shown below, where ΪΤ represents the number of times that it has been changed.

Ww3 = Nw3 / (Nwi + l) .W ^ + \ / (N ^ +1) .L〇

Nw3 = N ^ + l (11) (c) For the L4 layer nodes that are not the maximum value in the L4 layer, the output value is sent in the recall mode. Assume that the correlation value stored in its node is Fei, and Ji represents the number of times that it has been changed. The correction method is shown below, which is negative learning.

Wr = (Nr + b) l (Nr-> rb + \) * Wr + \! {Nr +

Nr = Nr + \ '(12) where' δ represents a bias', which avoids a correlation value in the first few learnings. If negative learning is encountered, its value will not be reduced too much at once. . (Case 2) When the output value of node RA (L5) is less than the critical value θ3, the following learning procedure is performed: (a) If node SA (L2) outputs the maximum similarity value for query word 11 (coding style q) If it is smaller than the critical value Θ !, a new L2 node wl is established, and the initial values of the coding style and the number of times it is changed are set to i, = 1, ..., η

Nw] = 1 (b) If the node SA (L2) is used to retrieve the component key 19 (code sample 28) This paper size applies the China National Standard (CNS) A4 specification (210X297 mm) ------- I -I-*----- --- 1 —. 抑 衣 _ I--------II I- I XT (Please read the back of the page first; intentions before filling out this page j Ministry of Economy Central Standards Bureau staff consumer cooperation print 420778 λ7 ___B7 V. Description of the invention (26) ~~~~~~ Formula k) The maximum similarity value output is smaller than the critical value θ |, then a new L2 node is established w2, the initial values of the coding style stored therein and the number of changed times are set as: / ^ 2 = 弋, i = 1, ..., η

Nw2 = 1 (Η) (c) Establish a new L4 node W3, and set the initial value of the correlation value and the number of times it is changed as:

Ww3 = 1.0 ^ 3 = 1 (15) According to the descriptions of the synonym program unit 1 and the personalized synonym adjustment unit 9 described above, the information (software component) acquisition system of this embodiment is known, and it can not only dynamically Adjust the structure in the synonym program, and also provide the ability to input fault tolerance. The information index program (3, 6) shows a functional block diagram of the information indexing program section 3 of this embodiment. As shown in the figure, the 'information indexing program section 3 includes an index key temporary storage section 301 and a software component indexing unit. 3〇3, according to the index key 13 sent by the synonym program unit 1, generates the software component ID 15 which actually corresponds to certain conditions. As mentioned above, in the tenth embodiment, the query item entered by the operator There are 4's, which are software component function (functión), associated medium (medium), software component type (file type), and system type \ ype), as shown in Figure 3. Among them, the input of only software component functions is set freely, and after processing by the synonym program section, several related index keywords 13 are generated. For other index items, ___ 29 ^ " € scale applies Towel ί House Logo (CNS) A4 Specification (2IGX 297 mm) '{Please read the precautions on the back before filling this page) Wave-， 1Τ Λ7 420778 —--------- B7 ^ _ V. Description of the invention (27) ~ (Please read the notes on the back before filling this page) is one of the keywords provided by the system. Select the index key 13 (only corresponding to the software component function item) found by the synonym program section 1 in order, and combine the keywords of the four query items into a group of actual query data to the index key. The word temporary storage section 301 is stored separately. Then 'the keywords stored in the index key temporary storage unit 301 are sent to the software component index unit 303 for comparison at the same time to find the software components that meet a certain correlation, and then the identification code of these software elements is 15 Submit 'completes the function of software component capture. The above processing will continue until all the index keywords found are searched. Fig. 7 is a schematic diagram showing the detailed structure of the information indexing program section 3 using the structure of a fuzzy neural network in this embodiment. The fuzzy neural network used to implement this information indexing program section 3 includes two layers, namely L6 and L7, corresponding to the index key temporary storage section 301 and the software component indexing unit 3 in FIG. 6, respectively. . The node functions of [6 layer and [7 layer] are described in detail below. L6 layer L6 layer is composed of m sub-layers, which are labeled as L61, L62, .... Mm; and each sub-layer has η nodes, such as the printed sub-layer L61 of the Employees' Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs. It includes n nodes from QJ to qJ, and the sub-layer L6m includes n nodes to Qnm and so on. Each sub-layer is used to store a keyword in each query keyword group. Therefore, m represents the keywords included in each query keyword group, and m is 4 in this embodiment; and η represents The maximum possible number of letters for a keyword. In this embodiment, n is 21. In other words, 'in this embodiment, the nodes Q! 1 to Q21 in the sub-layer L61 * 30 national standard f CNS in this standard A4 specification (210x 297 mm-^-420778 Λ7 B7 V. The description of the invention (28) is used To store a keyword about the function of the software component, and this keyword is sequentially set using the index keyword 13 found by the synonym program unit 1. The nodes Q, 2 to q212 in the sub-layer L62 are Used to store keywords related to the medium. Nodes Q! 3 to Q2 in the sublayer [63] are used to store keywords related to the software type. The sublayer L64 The nodes Q44 to q2 | 4 are used to store keywords related to the system type. In this embodiment, the latter two keywords are selected from the keywords preset by the system. L7 layer L7 layer Includes u nodes, labeled Ci, C2, & .....

Cu. The nodes of each L7 layer correspond to the software components set by the system. In each L7 layer node, a software component keyword used to describe the corresponding software component and an identification code possessed by the corresponding software component are stored. Among them, the software component keyword has keywords corresponding to 4 query terms, and the value is input by the L6 layer node when the L7 layer node is established. The identification code (ID) is used to indicate the location of the software component in the database or the access location on the Internet. Each L7 layer node will receive each keyword from the L6 layer node and compare it with the keywords of the software components stored in it. If the conditions are met, then the identification code corresponding to this software component can be added. Output. The action of the node with _L7 layer can be described by the following formula: α (ϋΑ) = Σ ^ (16) 31 This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X 297 mm) (Please read the Please fill in this page again for attention) '11 Printed by the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs 420778 V. Description of the invention (29 f ~ Λ7 B7 if Km ^ OandiTm ^ c / 0 ifKm = 0 (1-) otherwise (17) ( 18) (19) M (Q, C,) = 2 * (1 + (d (Q, C,) / Fdfe) -1 〇 (Q = (MC ') ifM general, [Null? Otherwise No. (16) The formula is used to calculate the distance between the input keyword group α (consisting of 4 keywords) and the software component keyword group C stored in the i-th node Q describing the first software component. In the embodiment, the bean includes 4 vectors (1 to m) 'Each vector includes 21 elements (1 to n), and the same' Ci also includes 4 vectors, each vector also includes 21 elements In addition, the Am value is the distance representing one of the vectors (one of 1), which is calculated by Equation (17). Equation (17) is the calculation of a vector and software components in the input keyword pan. The distance Am between the corresponding vectors in the key encoding style Ci. In particular, only the query term (vector) part where the input keyword 3 does have an input keyword is calculated, and for the query term part τ without a keyword input, the distance is Am is set to 0. With this processing, even if the query is performed when the operator cannot determine a certain keyword (for example, the system type cannot be determined) 'this embodiment can be processed smoothly, that is, blurring can be achieved The purpose of input extraction, which represents the weight of the m-th vector (ie, the query term) in the input keyword group Q, and also the importance of all the vectors in the input keyword group. Equation 18 is It is used to calculate the similarity value between the input keyword group g and the software component keyword group ζ of the original storage node Q. Its Yin, 32 paper standards are applicable to China National Standards (CNS) Λ4 specification (2! 0 X 297 mm) m — ^ — fr —I— —I rlr —-I u ^ i -¾, -'e (Please read the notes on the back before filling this page) Printed by Zhengong Consumer Cooperative, Central Bureau of Standards, Ministry of Economic Affairs 11 420ϊ78 Λ7 Β7 V. Description of the invention (30) ΪΜ 'and Fe' are fuzzy operators, which are used to describe the range covered by the attribution function of the distance between the input keyword-level bean and the software component keyword set stored in this node. And the sensitivity to distance. The similarity value 'calculated in Equation 18 can not only be provided to this L7 node to determine whether it can output the identification code of the corresponding software component, but also can be provided to the component sorting and filtering program. The unit 5 determines the order in which the extracted software components are sorted. Equation 19 is used to actually determine whether to retrieve the software components corresponding to this L7 node. In other words, if the similarity value of the input keyword group $ and the software component keyword group ζ describing the software component is coincident) exceeds a predetermined threshold Θ4, it can be decided to output the software component identification code LcCCi ), And also output the corresponding similarity value to the component sorting and filtering program unit 5 for subsequent processing. The following is a description of the setting of the threshold value 04 in each L7 node. When starting to use the information retrieval system of the present invention, the operator will first input the query words of each query item on the query day surface as shown in Figure 2 to index the software component database. Some query word input methods used in this embodiment are directly input by the operator freely, for example, the software component function 62 ', while others use the pull-down (p0p_up) option to let the operator choose, such as the associated media 64 , Software component type 66, and system type 68. In the software component function 62 using direct typing, the operator can fill in multiple query words, and the query words and the query words are separated by a separator. When the system processes the query item, it will separate each query word, process one at a time, and finally combine the results. This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm). --- ^- --- Clothing— (Please read the notes on the back before filling out this page)-Nail. Printed by the Consumers 'Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs 420778 Α7 Β7 Printed by the Consumers' Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs 31) ~ " to handle. However, the following pull-type option losers, because each option is almost independent and mutually exclusive, so only one option can be selected at a time. In addition, only the query words in the software component function 62 will be processed by the synonym program unit 1, and many keywords will be obtained; as for other query items that do not require synonym processing, the query words selected by the operator Set keywords directly. In the information indexing program, there are 3 keywords. Each time, from each of the 4 items in the search, a keyword is selected to form a group of keyword groups for component extraction processing. For all software components found by the information indexing program section 3, if all are selected by the operator at this time, three situations may occur. First, the number of software components retrieved is moderate, so the operator can immediately find a suitable software component. Second, the number of software components retrieved is too large for the operator to choose. Third, the number of software components retrieved is too small, and the operator cannot find a suitable software component. The component sorting and filtering program section 5 described later can adjust the threshold value 04 of each L7 layer node to solve the problems caused by the following two cases. When the number of software components is too large, the critical value Θ4 is adjusted. In other words, the critical value Θ4 ′ used to control whether or not the software components output is adjusted to increase the number of software components to reduce the number of software components output. Conversely, when the number of software components is too small, the threshold value θ4 needs to be lowered to allow the operator to choose from more software components. According to the above description of the information indexing program section 3, it is known that the information (software component) extraction system of this embodiment can perform software component extraction in a fuzzy manner for the query word entered by the operator. Paper size is applicable to China National Standards (CNS) Α4 size (2! 〇χ297mm) II ----- n I 1 ^ ----- I__Γ (Please read the notes on the back first and keep writing and writing) 420778 Λ7 ________B7 V. Description of Invention (32) Take. That is, even if the operator cannot determine the correct inner valley of a query item, this information extraction system can find all possible software components in the closest way. (5) In addition to using the critical value 04 to filter the number of software components extracted, the component sorting filtering unit 5 can also use the corresponding similarity value sent by the information indexing program unit 3 as described above. To determine the order of each software component. For & 'when the similarity value is greater', the entered query phrase and the keyword describing the software component are closer, so their priority is higher; relatively, when The smaller the similarity value, the greater the difference between the two, so its priority is lower. When the extracted software components are output for the operator to choose, the output can be determined according to the size of the similarity value The priority order allows the operator to quickly select the required software components. The personalized sorting and filtering adjustment section 7 can adjust the ordering mechanism and Filtering mechanism to adjust according to personal usage habits and retrieval preferences. Central Industry Bureau, Ministry of Economic Affairs, Weekly Consumption Cooperation, Du Yinhua --..--.----- < Please read the note on the back before filling in this page] According to the above, the software component extraction system of this embodiment has the following advantages: ~ 1. Can be used according to the operator's usage habits and personal preferences, flexibility Adjust synonym dictionary to facilitate query; 2. Can provide the ability to input fault tolerance; 3. Provide the performance and processing of uncertain information, that is, provide fuzzy check 35's 1-scale universal towel SS] family standard rate (CNS ) A4 specification (2 丨 GXW? Public trend Γ -------------- Λ7 420778 _______B7 V. Description of the invention (33) 'Inquiry method; 4. Can be based on the operator's retrieval preferences And usage habits, sorting the extracted software components and filtering the software components with less relevance to increase the convenience of the operator when selecting. Although the present invention is disclosed above in a preferred embodiment, it is not intended to be used. To limit the present invention, anyone skilled in the art can do some modifications and retouching without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be defined by the scope of the appended patents as follows: Standard. (谙 Please read the notes on the back before filling this page) Printed by the Consumer Cooperatives of the Central Bureau of Standards of the Ministry of Economic Affairs This paper size applies to China National Standard (CNS) A4 (210X297 mm)

Claims (1)

420778 A8 B8 C8 D8 6. Scope of patent application — '1. — An information retrieval system for querying the corresponding information components based on the entered query word, the information retrieval system includes a word definition program' for The query word 'searches for at least one index key having a synonymous relationship; an information indexing program for searching for at least one corresponding information element according to the index keys; and a component sorting program' for The searched-out information elements are sorted and filtered for selection. The synonym word program includes: an encoding unit for encoding the query word and generating a query word code. Style 'The query word encoding style corresponds to the characteristics of the query word; a similarity calculation section' is used to compare the query word encoding style with the system synonym encoding style generated by the plural system synonym encoding to generate a corresponding Similarity values' These similarity values are used to indicate the similarity between the query word and the synonyms of these systems. The words respectively correspond to plural system keywords; and an association section 'the similarity values according to the similarity calculation section' and the correlation values between the system synonyms and the system keywords' Compare with a predetermined threshold to 'select output from these system keywords as the index keywords. 2. The information retrieval system described in item 1 of the scope of patent application, wherein the information components are various types, functions, and related media, and software components applicable to various operating platforms. 3 _ The information retrieval system described in item 1 of the scope of patent application, wherein the similarity calculation unit includes: J ^ i nn —1- I 1 ^ 1 1 1 ^ 1- I ^ 11 -M3., V " (Please read the notes on the back before filling out this page) Printed by the Consumers' Cooperative of the Central Bureau of Standards of the Ministry of Economic Affairs—37 420778 AS B8 C8 D8 Printed by DuPont of the Bureau of Standards and Technology of the People's Republic of China. Please include multiple similarities in patent scope. The calculation units are respectively used to store the system synonym word encoding styles corresponding to the system synonym words, and receive the query word encoding style, and calculate the corresponding similarity value according to a predetermined similarity calculation rule. 4. The information retrieval system described in item I of the patent scope, wherein the association section includes: a plurality of association units for receiving similarity values generated by the similarity calculation section, and according to these The correlation value is compared with the predetermined threshold to determine whether the corresponding system key is used as the index key. 5. The information retrieval system described in item 丨 of the scope of patent application, wherein the Sx encoding. According to the query word, the number of each letter, the position where the letter first appears, and all the letters are generated to generate the corresponding query word encoding format D 6. The information acquisition system described in item 5 of the scope of patent application The synonym program section also includes a position amount offset calculation section, which uses the first occurrence position of each letter to calculate the position amount offset to determine whether there is an input error. When there is an input error, the corresponding Query word to compensate. 7_ —An information retrieval system 'for querying corresponding information components according to the input query word'. The information retrieval system includes a synonym program for searching at least one index having a synonymous relationship based on the query word. Keywords; an information indexing program to search for at least one information component according to the index keywords; and a component sorting and filtering program 'to sort and search the information components through II n n -nn Is, j I n--In TJ! -9 (Please read the precautions on the back before filling out this page) 38 Printing Policy of Employee Consumer Cooperatives, Central Standards Bureau, Ministry of Economic Affairs 4 ^ iv i / ¾ A8 B8 C8 ____US I 、 Applicable patent scope ~~ ~~~ Filter and output for selecting the required information components, which is characterized by the following: _ The synonym program includes: An encoding unit 'is used to encode the query word and the corresponding information component. Extracting a component keyword to generate a query word encoding style and a retrieval component keyword encoding style, the query word encoding style corresponding to the characteristics of the query word, and the retrieval component key The word encoding style corresponds to the characteristics of the keyword of the retrieval component; a similarity calculation unit for separately identifying the query word encoding style and the retrieval component keyword encoding style with the system synonym generated by the plural system synonym encoding The coding styles are compared to generate the corresponding first set of similarity values and the second set of similarity values, which respectively represent the similarity between the query word and the retrieval component keyword and the synonymous words of these systems. The equivalent system synonyms are respectively corresponding to plural system keywords; an association unit 'according to the first group of similarity values and the second group of similarity values generated by the similarity calculation unit, and the system synonyms and the Calculate the corresponding output value based on the original correlation value between system keywords; and a judgment unit, based on the first group similarity value, the second group similarity value, and the output values, determine whether to modify the similarity calculation Department and the associated department. 8_ The information retrieval system described in item 7 of the scope of the patent application, wherein the information retrieval system further includes a synonym adjustment unit, and adjusts the similarity calculation unit and the associated unit according to the judgment result of the judgment unit. . 9_ If the information retrieval system described in item 7 or 8 of the scope of patent application, 39 units --- (Please read the precautions on the back before filling out this page) CNS in this paper size) (21GX297 company Li plant '420778 Μ B8 C8 D8 VI' patent application scope, of which 70 pieces of information are various types, functions, and related media, as well as software components applicable to various operating platforms. 10. If the scope of patent application is 7 or 8 In the information retrieval system, the similarity calculation unit includes: a plurality of similarity calculation units respectively configured to store system synonym word coding styles corresponding to the system synonym words, and respectively receive the query word code style and the query word code style Retrieve the component keyword encoding style to calculate the first set of similarity values and the second set of similarity values. 11. The information retrieval system described in item 8 of the scope of patent application, wherein the judgment section includes- A first selection unit for selecting the first town with the largest similarity and the second group with the largest similarity, respectively, and comparing it with a first predetermined threshold to determine the similarity meter 12. The information retrieval system as described in item 8 of the scope of patent application, wherein the judgment section includes a second selection section for selecting the largest of these output values and connecting with a first Two predetermined thresholds are compared to determine the similarity calculation section and the related section. 13_ The information retrieval system described in item 7 or 8 of the scope of patent application, wherein the encoding section is based on the query word and the Extract component keywords, record the number of all letters, the first occurrence of the letters, and all the letters themselves. -14. The information retrieval system described in item g of the patent application scope, wherein the synonym program department is still Contains a position amount offset calculation section, which uses the relative position of the first occurrence of each letter to calculate its position amount offset. 40 paper sizes are applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) --- -^ ------- ^ -— (Please read the notes on the back before filling out this page) -Order _ Printed by Employee Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs Α8 Β8 C8 D8 4 ^ 0ϊ78 6. Apply for a patent Range 'when If you make a mistake, you will be compensated to determine whether you have entered a wrong query. „------ ^ ------- ΐτ (Please read the precautions on the back before filling this page) 15_ — An information retrieval system for querying corresponding information components based on the input query word. The information retrieval system includes a synonym program for searching for at least synonymous relationships based on the query word. — Index key Words;-information indexing program 'to search for at least components based on these index keywords; and-component sorting and filtering program to sort and filter the information components found and output for selection The required information components are characterized in that the information indexing process includes:-an index key temporary storage unit that sequentially stores the index keys; and a plurality of element index units corresponding to the information elements' and storing the The component keyword groups of these information components, and the corresponding information 7G component identification codes, are used to calculate the relationship with the stored keyword when the input index keyword group is received. Similarity between members keyword group, and determines whether more than one predetermined threshold value to determine whether the output code corresponding to the identification information element. Printed by the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs 16. The information retrieval system described in item 15 of the scope of patent application, where the information components are various types, functions, and related media, and software suitable for various operating platforms element. 17. The information retrieval system described in item 5 of the patent application scope, wherein the predetermined threshold value is adjustable 'to determine the number of information element identification codes that can output corresponding element index units. 41 Qiao · Zhang Zhili) ---- 4 * :) A8 B8 C8 D8 Printed holes and applied for patents by the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs 18. A type of information retrieval system, based on the input query words To query the corresponding information element; it includes: a synonym program unit which uses the query word to search for at least one index keyword having a synonymous relationship according to a synonym fuzzy mechanism; an information index program unit is used to According to the index keywords, a corresponding at least one information component is searched out. A component sorting and filtering program unit sorts and filters the searched information components according to a sorting and filtering fuzzy mechanism, and outputs it for selection. Required information components; a humanized synonym adjustment unit that adjusts the synonym fuzzy mechanism in the synonym program unit according to the selected information element; and a humanized sorting filter adjustment unit based on the selected information element Information element, adjusting the order filtering fuzzy mechanism of the order filtering process part of the element. · 19. The information retrieval system described in item 18 of the scope of patent application, wherein the information components are 'types, functions, and related media, and software components applicable to various operating platforms. 20. The information retrieval system described in item 8 of the patent application, wherein the synonym program department is composed of a type of neural network to achieve parallel and fast processing. 21: The information retrieval system described in item 18 of the scope of the patent application, wherein the information / programming department is composed of a class-like neural network for parallel fan processing. 22. The information retrieval system as described in item 18 of the scope of patent application, which (锖 read the note on the back ^ > then fill out this page) -1 pack- 42C778 A8 B8 C8 D8 6. The synonymous subroutine unit in the scope of patent application includes a coding unit to encode the query word. The coding format of the query word includes the number of all letters in the query word and the first occurrence of the letter Position, and all the letters themselves; and a position offset calculation unit, which uses the position of the first occurrence of each letter to calculate its position offset to determine whether there is an input error. When there is a rotation error, then Compensate the query word. {Please read the notes on the back before filling this page) -3 Printed by the Consumer Cooperatives of the Central Bureau of Standards of the Ministry of Economic Affairs This paper applies the Chinese National Standard (CNS) A4 specification (210X297 mm)