JP6712521B2 - Search formula presentation system, search formula presentation method, and program - Google Patents

Description

The present invention relates to a technique of presenting a search formula used for searching a publication such as a patent publication (including an application publication publication, a patent publication publication, and a republication) to a user.

When searching for publications such as patent gazettes, the user specifies search items such as “applicant”, “inventor”, “claim”, and “full text”, and inputs a corresponding search keyword. To do. A combination of search items and search keywords is called a search expression. It is also possible to construct a search expression by further combining such search expressions.

Patent Document 1 discloses a technique for correcting an input search expression. In Patent Document 1, a screen including candidate keywords that are keywords related to the search keyword included in the search formula is displayed. Then, the user specifies a candidate keyword according to this screen, and the modified search formula is constructed.

JP2012-83935A

The user could not confirm which part of the search formula before correction was corrected from the corrected search formula. That is, the user could not confirm the difference between the modified search expression and the unmodified search expression.

A search expression presentation system according to an embodiment of the present invention includes a reception unit that receives a first search expression input at a user terminal, and a second search expression that is obtained by modifying the first search expression that has received the input. And a output unit that outputs the second search expression to the user terminal so that the location where the correction is performed is identifiable on the user interface screen of the user terminal.

Here, the search target is not limited to patent gazettes, and may be articles, newspapers, magazines, and all that can be searched using a search formula. Also, not only Japanese documents but also foreign documents such as English are targeted.

According to the present invention, the user can confirm the corrected part in the search formula.

It is a figure showing an example of composition including a search type presentation system. It is a figure which shows an example of the user interface screen displayed on a user terminal. It is a figure which shows an example of the user interface screen displayed on a user terminal. It is a figure which shows an example of the user interface screen displayed on a user terminal. It is a figure which shows an example of the user interface screen displayed on a user terminal. It is a figure which shows an example of the user interface screen displayed on a user terminal. It is a figure which shows an example of the user interface screen displayed on a user terminal. It is a figure which shows an example of the user interface screen displayed on a user terminal. It is a figure which shows an example of the search formula presentation method performed with a search formula presentation system. It is a figure which shows an example of the layout of the user interface screen displayed on a user terminal. It is a block diagram of a search type presentation system including detailed composition of a learning model engine. It is a figure which shows an example of a structure of a learning model. It is a figure which shows the other example of the search type presentation method performed with a search type presentation system.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The configurations described in the following embodiments are merely examples, and the present invention is not limited to the illustrated configurations.

<Embodiment 1>
<Structure>
FIG. 1 is a diagram showing an example of a configuration including a search expression presentation system 100 according to the embodiment. The search expression presentation system 100 includes a reception unit 110, a correction unit 120, an output unit 130, a correction engine unit 140, a control unit 150, and a search processing unit 160. The correction engine unit 140 includes an expert system engine 141, a thesaurus dictionary engine 142, and a learning model engine 143. FIG. 1 shows only an example of the configuration, and may include other configurations. In addition, not all of the configurations illustrated in FIG. 1 are indispensable requirements.

The search expression presentation system 100 can be realized as an information processing device. The information processing device may include a CPU, a memory, a HDD, and a network interface. In each unit shown in FIG. 1, the CPU may function as each unit shown in FIG. 1 by temporarily reading the program stored in the HDD into the memory and causing the CPU to execute the program read into the memory. .. Further, at least a part of each unit shown in FIG. 1 may be realized by a plurality of information processing devices mutually connected through various networks. Further, at least a part of each unit illustrated in FIG. 1 may be realized by distributed processing by a plurality of information processing devices. Each engine included in the correction engine unit 140 is a processing unit that performs a unique process. Each engine may be configured as an information processing device as described above. The configuration may be such that a plurality of engines are included in one information processing device. Alternatively, some engines (for example, the learning model engine 143) may be realized by distributed processing by a plurality of information processing devices.

The search expression presentation system 100 is configured to be capable of communicating with a terminal used by a user (hereinafter referred to as a user terminal 10) through various networks (Internet, LAN, WAN, etc.). The search formula presentation system 100 outputs the user interface screen used for the search to the user terminal 10. The search formula presentation system 100 can accept the input of the search formula from the user terminal 10 through the user interface screen and present the corrected search formula. In this embodiment, the user interface screen used for the search will be described by taking the form output from the output unit 130 as an example (as one implementation method of this form, provided by ASP (Application Service Provider)). Form), but is not limited to this. For example, the user terminal 10 may have previously installed an application corresponding to the present search formula presentation system. In this case, the user interface screen used for the search may be displayed on the user terminal 10 by the application installed on the user terminal 10.

The user terminal 10 may be any type of terminal such as a personal computer, a tablet, or a mobile terminal. In the present embodiment, the user terminal 10 has an input unit 11 for inputting a search expression and a display unit 12 for displaying a user interface screen including the search expression. The user terminal 10 may have functions other than the input unit 11 and the display unit 12. The search formula presentation system 100 manages the login of the user, and can also modify the search formula unique to the logged-in user. Details will be described later.

The receiving unit 110 receives the first search expression input from the user terminal 10. For example, on the user interface screen displayed on the display unit 12, the user inputs the first search expression from the input unit 11. Then, the input search formula is sent from the user terminal 10 to the search formula presentation system 100. The reception unit 110 receives the first search expression input at the user terminal 10. In this way, the reception unit 110 can receive another operation instruction from the user terminal 10.

The correction unit 120 corrects the first search expression that the input from the user terminal 10 is received in the reception unit 110, and generates the second search expression. Specifically, the modification unit 120 modifies the first search expression using at least one of the plurality of modification engines included in the modification engine unit 140. In the example of FIG. 1, an expert system engine 141, a thesaurus dictionary engine 142, and a learning model engine 143 are included as a plurality of correction engines. Details of the process for correcting the search formula will be described later.

The output unit 130 outputs the second search expression generated by the correction unit 120 to the user terminal 10 so that the corrected portion is displayed in a distinguishable manner on the user interface screen. For example, on the user terminal 10, the modified portion is highlighted on the user interface screen.

The control unit 150 performs control in accordance with an operation instruction from the user terminal 10 with respect to the location where the second search formula is modified. For example, control is performed to reflect the corrected content in the learning model used in the learning model engine 143. Details will be described later.

The search processing unit 160 performs a process of searching the publication database using the specified search formula and acquiring the search result.

<Example of user interface screen>
Next, a specific example will be described using an example of the user interface screen.

FIG. 2 is a diagram showing an example of a user interface screen 200 displayed on the user terminal 10. In this example, the user specifies “applicant/right holder” as the search item of the first search formula 210 and inputs “Taro Yamada” as the search keyword.

At this time, when the user presses the search button 280, normal search processing is performed. That is, the first search formula 210 is transmitted from the user terminal 10 to the search formula presentation system 100, and is accepted by the acceptance unit 110. The reception unit 110 sends the first search expression to the search processing unit 160, and the result searched by the search processing unit 160 is output to the user terminal 10 by the output unit 130. In the example of FIG. 2, this first search formula 210 indicates that the search result is 0. At this time, the user who notices that the input items of the first search formula 210 are incorrect corrects the search formula himself and presses the search button 280 again. As a result, the search result based on the modified search formula is output.

On the other hand, for example, a user who is not skilled in searching may not immediately understand what is wrong with the first search formula 210. In addition, there is a case where it is not possible to immediately judge how to correct it if it is wrong. In the present embodiment, the user presses the correction button 220 while the first search formula 210 is being input as shown in FIG. From the user's point of view, by simply pressing the correction button 220, the second search expression obtained as a result of the correction of the first search expression 210 can be confirmed. This will be described below.

When the correction button 220 is pressed while the first search formula 210 is being input, the first search formula 210 is transmitted from the user terminal 10 to the search formula presentation system 100. The receiving unit 110 receives the input of the first search expression 210. The modification unit 120 sends the first search expression 210 to at least one modification engine included in the modification engine unit 140. In the present embodiment, the correction unit 120 sends the first search expression 210 to each of the expert system engine 141, the thesaurus dictionary engine 142, and the learning model engine 143. The modification unit 120 receives the content modified by each engine and reconstructs the content in the form of a search expression. That is, the correction unit 120 generates the second search expression based on the content corrected by each engine. In the second search expression, the location where the modification is performed is associated with, for example, a modification flag indicating the type of modification engine and the modification content. That is, the correction unit 120 can associate a correction flag indicating what kind of correction was made by which correction engine with the place where the correction was made. The output unit 130 outputs a user interface screen including the second search expression to the user terminal 10. Note that, as described above, the output unit 130 may send the second search expression to the user terminal 10 instead of sending the user interface screen including the second search expression.

FIG. 3 is a diagram showing an example of the user interface screen 300 displayed on the user terminal 10 after the correction button 220 is pressed in the state of FIG. In FIG. 3, the first search expression 210 shown in FIG. 2 is continuously displayed. Further, in FIG. 3, the second search formula 320 generated by the correction by the correction unit 120 is displayed. In this example, the two search expressions 330 and 340 are generated by modifying the first search expression 210. The second search formula 320 is a general term for the generated search formulas 330 and 340. That is, the numbers of the first search formula and the second search formula do not have to match. In the present embodiment, the first search formula 210 and the second search formula 320 are both displayed in the user interface screen 300. According to such a configuration, the user can compare and confirm the search formula input by the user and the search formula corrected (corrected) by the search formula presentation system. In this way, since the correction result of the search formula can be confirmed, it can be used, for example, for the purpose of improving the search skill of the user.

In the user interface screen 300 of FIG. 3, the part of the second search formula 320 that has been corrected by the correction unit 120 is highlighted. In this example, the background color of the corrected portion is made different from that of the uncorrected portion to emphasize the corrected portion. Further, in the example of FIG. 3, the background color of the corrected portion is used properly so that the corrected portion and the correction content can be visually recognized. However, it is not limited to this example. For example, the characters such as color, size, font, and underline may be changed, or the background color and pattern of the characters may be changed. Further, instead of changing the corrected portion, the corrected portion may be changed to display the corrected portion in a distinguishable manner. With such a configuration, the user can confirm the portion of the second search formula 320 that has been corrected. Further, in the example of FIG. 3, the first search formula 210 and the second search formula 320 are shown together, but also in the form in which only the second search formula 320 is displayed. The user can confirm the place where the correction was made.

In the present embodiment, the second search formula 320 is arranged on the user interface screen 300 output from the output unit 130 to the user terminal 10, and the part of the second search formula 320 associated with the correction flag is The display mode may be changed according to the content of the correction flag. Then, the user interface screen 300 including the second search formula 320 whose display mode is changed in this way is transmitted from the output unit 130 to the user terminal 10, and the user interface screen 300 may be displayed on the user terminal 10. In addition, in a form in which the user interface screen 300 itself is not transmitted from the search formula presentation system 100, the second search formula associated with the correction flag is transmitted from the output unit 130 to the user terminal 10, and the search of the user terminal 10 is performed. An application corresponding to the expression presentation system 100 may change the display mode based on the correction flag. Here, the mode using the correction flag has been described as an example, but the present invention is not limited to this. Any method may be adopted as long as the correction location and the correction content can be specified.

In the example of FIG. 3, it is clearly indicated that the search item 331 in the search formula 330 has been modified. It is clearly indicated that the search item 341 in the search formula 340 and the search keywords 342, 343, 344, 345, 346 have been corrected.

Further, in the example of FIG. 3, index panels 351 to 357 indicating the type of the correction engine and the correction content are displayed. By referring to the index panels 351 to 357, the user can confirm what kind of correction was made by any one of the expert system engine 141, thesaurus dictionary engine 142, and the learning model engine 143. it can. The index panel 351 indicates that the item has been added by the thesaurus dictionary engine 142. The index panel 352 indicates that the item has been added by the expert system engine 141. The index panel 353 indicates that the learning model engine 143 has added the item. The index panel 354 indicates that the thesaurus dictionary engine 142 has changed the item. The index panel 355 shows that the expert system engine 141 has changed the item. The index panel 356 indicates that the learning model engine 143 has changed the item. Index panel 357 indicates that the expert system engine 141 has deleted the item. In the present embodiment, the color of the index panel 356 and the color clearly indicated as the correction portion match, and the user indicates to the index panel 351 to 357 what kind of correction engine has made what kind of correction. You can confirm it by referring.

In the example of FIG. 3, the form using the index panels 351 to 357 will be described as an example, but the present invention is not limited to this. For example, without displaying the index panels 351 to 357 on the user terminal 10, the type of the correction engine and the correction content may be displayed when the cursor is over the correction portion. Alternatively, the type of the correction engine and the correction content may be displayed by displaying a balloon from the correction location. Further, although the example of FIG. 3 shows an example in which the display form is changed, the present invention is not limited to this. A sound may be output when the cursor is over the correction point, and the output sound may be changed according to the type of the correction engine and the correction content. Alternatively, a voice output function may be provided so that the type of the correction engine and the correction content are output by voice when the cursor is over the correction portion.

According to such processing, the user can confirm the type of the correction engine and the correction content. Therefore, the user can easily determine whether to apply the correction as it is to perform the search, to perform the search after further modifying, or to not apply the modification. For example, if the user already has knowledge that the reliability of a specific correction engine is high or low for a specific search item, the user can quickly determine the accuracy of the correction made by the correction engine. It becomes possible to do.

Further, in the case where a plurality of correction engines are included in a specific type of correction engine, the information for identifying the correction engine may be specified. For example, when the learning model engine 143 includes a plurality of types of learning models, information for identifying the learning models may be specified.

Further, as shown in FIG. 3, in the present embodiment, when a correction location is selected by, for example, right-clicking, menus 361 to 365 of various operation instructions are displayed. When a part of the menus 361 to 365 is selected (input), an operation instruction is sent from the user terminal 10 to the reception unit 110. The control unit 150 performs control based on this operation instruction.

The menu 361 is an instruction for the user to edit and change the contents of the corrected portion. The menu 362 is an instruction to delete the contents of the corrected portion. When these menus are selected, further modifications are made to the modified search expression.

The menu 363 is an instruction to search the document list including the keyword of the corrected portion. When this instruction is accepted by the acceptance unit 110, the control unit 150 causes the search processing unit 160 to perform a search using the instructed search keyword. The search result is presented to the user terminal 10 by the output unit 130.

The menu 364 is an instruction for causing the artificial intelligence (learning model) to learn the content of the corrected portion. When this instruction is accepted by the accepting unit 110, the control unit 150 causes the learning model included in the learning model engine 143 to learn the content of the corrected portion. Details will be described later.

The menu 365 is an instruction to display thesaurus terms (synonyms) related to the keyword of the corrected portion. When this instruction is received by the receiving unit 110, the control unit 150 inputs the keyword of the correction location into the thesaurus dictionary engine 142, transmits the result obtained from the thesaurus dictionary engine 142 to the output unit 130, and the output unit 130 causes the output unit 130 to output the result. The result is output to the user terminal 10.

Next, each correction engine will be described. The expert system engine 141 is a program composed of a rule group for analyzing information on a specific field. The expert system engine 141 has, for example, an inference engine and a knowledge database. The expert system engine 141 in the present embodiment has a knowledge database that stores various know-how in the search service, and the inference engine corrects based on the knowledge database. For example, "Applicant/Rightholder" is often a company name, and if "Device" is simply entered as a search keyword in "Claims", it makes little sense to search. , It has a knowledge database that stores various know-how. The expert system engine 141 makes corrections based on such a knowledge database.

The synonym dictionary engine 142 has a synonym dictionary database, and makes corrections based on synonyms related to the input term.

The learning model engine 143 has a plurality of learning models corresponding to the search items, and inputs the search keyword into the learning model corresponding to the search items of the search formula to output the output data. The correction unit 120 performs correction based on the output data.

A specific example of FIG. 3 will be described. As described above, in the first search formula 210, “applicant/right holder” is specified as the search item, and the personal name “Taro Yamada” is input as the search keyword. It is known that the “applicant/right holder” is often a company name and the individual name is often used as an inventor. Based on the knowledge database having such knowledge, the expert system engine 141 outputs the result of modifying the search item from “applicant/right holder” to “inventor (latest)”. In response to this, the correction unit 120 generates the search formula 330. The output unit 130 outputs to the user terminal 10 the search formula 330 that clearly indicates that the expert system engine 141 has changed the search formula 330. The output unit 130 inquires of the search processing unit 160 about the number of search results when the search formula 330 is used, and also outputs the result to the user terminal 10.

It is also known that personal names have a high probability of having the same sex and same name. Therefore, the expert system engine 141 extracts and presents the applicants/rights holders including the inventor “Taro Yamada” as shown in the search formula 340. As a result, the user can easily reach the desired search result. Then, for example, the user can right-click on the unnecessary applicant/right holder and select the menu 362 to delete the unnecessary applicant/right holder from the search formula 340.

Check boxes 370 to 372 are buttons for the user to instruct whether or not to perform the search by applying the corrected second search expression. When the check box 370 is checked, both the check boxes 371 and 372 are checked. When the check box 371 or the check box 372 is checked and the search button 380 is pressed by the user, the instruction to execute the search is received by the reception unit 110, and the search processing unit 160 causes the checked search expression (second search). The search is performed according to the formula), and the output unit 130 outputs the search result to the user terminal 10.

As described above, according to the processing of the present embodiment, the portion of the second search formula that has been corrected is displayed in an identifiable manner. Therefore, the user can easily confirm the corrected portion. Further, according to the processing of the present embodiment, the type of the correction engine and the correction content are clearly displayed. Therefore, the user can refer to the type of the correction engine and the content of the correction and instruct further action to the corrected portion.

In the example explained above, "Taro Yamada" is entered and "Applicant/Rightholder" is specified as the search item, but it is assumed that this search item is not specified. Even if it is supposed, according to the present embodiment, “applicant/right holder” can be designated as a search item.

FIG. 4 is a diagram showing a user interface screen 400 when another search formula is used. On the user interface screen 400, a first search formula 410 and a second search formula 450 generated by modifying the first search formula 410 are displayed.

The modified part 451 is a part added by the learning model engine 143. For example, the learning model engine 143 can utilize a large amount of patent data as a context vocabulary and correct misspellings based on contextual usage in the patent specification. Here, the learning model engine 143 corrects the misspelling of the word canceler to the word cancer (adding the word cancer to the search formula). Such a misspelled word is often not found by the thesaurus dictionary engine 142. Although the example in which the learning model engine 143 corrects is described here, when the synonym dictionary engine 142 has a database that stores data on synonyms including misspellings regarding the word canser, the synonym dictionary engine 142 uses the synonym dictionary engine 142. You may modify and add the word cancer.

The modified part 452 is a part (search formula) deleted by the expert system engine 141. In the example of FIG. 4, the same search keyword is used to generate a first search expression 410 including a search expression in which “all items (full text)” are specified as search items and a search expression in which “summary” is specified. It has been entered. It is known that "all items (full text)" include "summary". Therefore, the expert system engine 141 deletes the search expression in which "summary" is specified as the search item.

The operation instruction 461 is an instruction to translate the English word into a Japanese dictionary. The operation instruction 462 is an instruction to restore the deleted search expression.

FIG. 5 is a diagram showing a user interface screen 500 when another search expression is used. The user interface screen 500 displays a first search expression 510 and a second search expression 550 generated by modifying the first search expression 510. The first search expression 510 is an example in which “inventor (latest)” is specified as the search item and “Nomura Research Institute” is input as the search keyword. In this example, the user actually wants to search for a document in which the “applicant/right holder” is “Nomura Research Institute, Ltd.”, but the input is made as in the first search expression 510. It is an example of the case of being broken.

If the company name is omitted, it may not match the company name thesaurus, so the synonym dictionary engine 142 may not correct it. Further, the expert system engine 141 may recognize that the personal name is “Nomura Research Institute” and may not correct it. On the other hand, the learning model engine 143 may be able to make a correction. For example, a learning model in which learning is performed using a large amount of data of publications is prepared. In this learning model, it is assumed that the feature that the word “Nomura Research Institute” and the word “Nomura Research Institute, Ltd.” are related has already been learned. In this case, by inputting the information of the first search expression 510 to the learning model engine 143, “Nomura Research Institute” is corrected to “Nomura Research Institute, Ltd.” as indicated by the correction point 552. Further, the result of the correction made by the correction engine may be used to make another correction by another correction engine. In the example of FIG. 5, as a result of the learning model engine 143 modifying “Nomura Research Institute, Ltd.”, the expert system engine 141 sets the search item to “applicant/right holder (latest )” has been corrected. In this way, the second correction engine may make further corrections based on the result corrected by the first correction engine.

There are a plurality of learning model construction methods even in supervised learning, and there is no intention to limit the construction method of the present invention to a particular method, but an example will be described. In the above example, the input value “inventor (latest)” “Nomura Research Institute” and the output values “applicant/right holder” “Nomura Research Institute, Ltd.” are trained by the neural network as supervised data. As another example, the input value is a specific name (name, company name, etc.), and the output value is a search item name. It is also possible to construct a learning model α that outputs which search item the value corresponds to. In addition to this, countless supervised data whose input values are incorrect concrete names or names that are not official names (name, company name, etc.) and output values are names that are official names (name, company, etc.) It is also possible to construct a learning model β that prepares and learns, and outputs an official name when an abbreviated name is input. By using these learning models α and β and inputting “Nomura Research Institute” into the learning model β, “Nomura Research Institute, Ltd.” is output, and the output “Nomura Research Institute, Ltd.” is learned model α. It is also possible to realize the modification described above by inputting into "" to output "applicant/right holder".

Further, although a search expert often sets a synonym for a certain search term in a search expression, a search beginner may perform a search without using a synonym. Therefore, a learning model that outputs synonyms can be constructed. As an example of constructing a learning model for synonyms, a search formula of a search expert or a search formula of an examiner of the Patent Office can be used. More specifically, among these search formulas (search formula example: A* B*(C+C'+C'')), part of the search expression that is configured using the parameters of the OR logic operation (part of the search expression: C+C'+C'') , Create supervised data using a part of the retrieved retrieval formula (example of supervised data: supervised data 1 (input value C, output value C+C'+C''), supervised data 2 ( Input value C', output value C+C'+C'')), supervised data 3 (input value C'', output value C+C'+C'')) , C', and C'' are input by the user, C+C'+C'' is generated as the output value.

FIG. 6 is a diagram showing a user interface screen 600 using the same search formula as in FIG. On the user interface screen 600, a first search formula 610 and a second search formula 650 generated by modifying the first search formula 610 are displayed. In the example of FIG. 6, in addition to the presentation of the second search formula 650, the recommendation message 660 is included in the user interface screen 600. The recommendation message 660 indicates recommended search conditions that are not included in the second search expression 650. The output unit 130 outputs a classification code of a frequently used field as a recommendation message 660 based on past search history information of a user who has logged in using the user terminal 10. In the recommendation message 660 of FIG. 6, the classification code of the recommended field is presented based on the output data obtained from the learning model in which the user's past search history is learned.

FIG. 7 is a diagram showing another user interface screen 700. Similar to FIG. 2, FIG. 7 shows a screen on which the user arbitrarily inputs a search expression (first search expression 710). Further, the user interface screen 700 includes an item 750 for inputting a target number of hits, which is the desired number of search results. Some users have a desire to construct a search formula so that a specific number of publications are extracted, for example, a list of 200 publications is desired. Here, the number of search results obtained when the user inputs the first search formula 710 and presses the search button 780 may not be the desired number (200 in this example). At this time, the user specifies “200” in the item 750 for inputting the number of target search results, and presses the correction button 720. Then, the reception unit 110 receives the number of the specified target search results. The correction unit 120 corrects the search expression so that the value is close to the number of target search results (when the search expression is corrected to create a population close to the number of target search results, each search item is added to the search item). Although it is possible to correct the input value (such as changing the search period), it is also possible to modify the logical expression that combines a plurality of search item expressions shown in Fig. 7. In general, the number of records that can be obtained is very large, so providing a logical formula that is closest to the number of target search results is one example. , The configuration may be such that a plurality of logical expressions are presented to the user in the order close to the number of cases and selected.

The correction unit 120 may correct the number of target search results to be equal to or less than the number of target search results, or may be corrected to be within a predetermined range from the number of target search results. For example, when the target search result is 200, the correction unit 120 may correct the search expression so that the number of search results is 200 or less, or the number of search results is 180 or more and 220 or less. The search formula may be modified so that The correction unit 120 approaches the number of target search results when the number of search results of the second search formula generated by correcting the first search formula 710 is not within the predetermined range from the number of target search results. Thus, the second search expression can be further modified to generate the third search expression. Here, if the number of target search results is extracted and there is no desired publication after screening the extracted publications, the third search formula is used so as to additionally extract the publications of the number desired by the user. It can be modified to generate new search expressions. That is, a new search expression is generated so that the number of publications screened from the publications extracted by the new search expression (the number of cases extracted by the third search expression) is removed to be the number desired by the user. ..

FIG. 8 is a diagram showing another user interface screen 800. The user interface screen 800 displays a first search formula 810 and second search formulas 850 and 860 generated by modifying the first search formula 810. FIG. 8 shows an example in which a plurality of correction modes are prepared and the second search expressions 850 and 860 corresponding to the correction modes are displayed. In the example of FIG. 8, the intellectual property department mode and the examiner mode are prepared. Such a plurality of correction modes can be realized by building a learning model used by the learning model engine 143 as a learning model corresponding to each mode and switching the learning model used by the learning model engine 143. For example, past search formulas for a specific group or a specific user (search formula information for examiners of the JPO (search formulas for each examiner) are stored in the patent bag, and search formula information for the IP Department is Since it is recorded on the patent search system, it is possible to obtain each. Not limited to the Japan Patent Office, it may be searchable information of examiners of overseas patent offices. A search model is registered using the search formula). In the example of FIG. 8, a learning model that outputs a search formula in the intellectual property department mode and a learning model that outputs a search formula in the examiner mode are respectively constructed. The learning model engine 143 inputs the first search expression for each learning model and obtains output data from the learning model. The correction unit 120 generates each second search formula based on the output data output from each learning model. In the example of FIG. 8, a second search formula 850 based on a search formula frequently used by a person in charge of the intellectual property department in a search and a second search formula 860 based on a search formula frequently used by an examiner in examination are included. A user interface screen 800 is displayed. It should be noted that the learning model for each specific person in the intellectual property department and the learning model for each examiner may be further subdivided, and in this case, the user designates a specific intellectual property person or examiner and corrects it. Instruct processing.

Further, such a plurality of correction modes may be used together with the number of target search results described in FIG. 7. For example, if the number of target search results is set to 200, and if the result of the second search formula is larger than 200, even if the correction mode is selected as a guideline for deciding how to narrow down, Good. The example of FIG. 8 indicates that the number of cases can be narrowed down to 200 cases or less by the correction location 854. Although not shown in FIG. 8, the search formula for narrowing down may be different (may be the same) in the intellectual property department mode and the examiner mode. It should be noted that the corrected portions 851, 852, 861, and 862 in FIG. 8 indicate the portions added by the thesaurus dictionary engine 142.

<Flowchart>
FIG. 9 is a flowchart showing an example of a search formula presentation method performed by the search formula presentation system according to the present embodiment. The process of FIG. 9 shows a process after the user logs in to the search formula presentation system 100 using the user terminal 10 and a request for displaying a search screen is transmitted from the user terminal 10 to the search formula presentation system 100. There is.

In step S901, the output unit 130 presents the user interface screen used for the search to the user terminal 10. The user terminal 10 displays the user interface screen on the display unit 12 and waits for the input of the first search expression from the user.

When the first search formula is input in the user terminal 10 and the correction request button for the first search formula is pressed, the reception unit 110 inputs the first search formula and the correction request in step S902. Accept. The accepted first search formula is transmitted to the correction unit 120.

In step S903, the correction unit 120 transmits the first search expression to at least one of the plurality of correction engines. In this embodiment, the information is transmitted to the expert system engine 141, the thesaurus dictionary engine 142, and the learning model engine 143.

In step S904, the correction unit 120 identifies the item corrected by the correction engine. That is, the corrected search item and search keyword are specified in the first search formula.

In step S905, the correction unit 120 generates a second search formula that reflects the corrected items as a search formula. In addition, a correction flag indicating the content of the correction is associated with the correction location.

In step S906, the output unit 130 presents the user interface screen including the first search formula and the second search formula to the user terminal 10 in such a manner that the correction location can be identified. For example, in the second search formula, the corrected portion is highlighted and displayed on the display unit 12 of the user terminal 10.

As described above, according to the present embodiment, the location of the modified search formula is displayed in a distinguishable manner on the user interface screen used for the search. As a result, the user can confirm which part of the search expression initially input has been modified. Further, according to the present embodiment, the search formula input by the user and the modified search formula are displayed together. The user can compare and confirm the search formula input by the user and the search formula corrected (corrected) by the search formula presentation system.

FIG. 10 is a diagram showing an example of the layout of the user interface screen 1000. In the examples described so far, the form in which the search formula before the correction and the search formula after the correction are separately displayed at the top and bottom of the screen has been described. However, the screen layout is not limited to this example. For example, as shown in the user interface screen 1000 of FIG. 10, a search formula before correction and a search formula after correction may be displayed on the left and right sides of the screen. Area 1010 is an area in which the first search expression is displayed, and area 1020 is an area in which the second search expression is displayed. By displaying the left and right in this way, it becomes easier for the user to compare the search expression before correction with the search expression after correction. The layout can be changed by the user, and the search formula before correction and the search formula after correction can be arranged at the top and bottom of the screen or on the left and right of the screen. You can also Further, the layout of the search formula before the correction and the search formula after the correction may be arranged in the opposite layout to the above-described example. Further, when the display unit 12 of the user terminal 10 is small, such as a smartphone, a switching button is displayed, and each time the user presses the switching button, a search formula before correction and a search formula after correction are displayed. May be switched and displayed on the display unit 12 of the user terminal 10.

<Embodiment 2>
In the second embodiment, a mode in which the learning model engine 143 of the correction engines described in the first embodiment is used will be described in detail.

<Structure>
The basic configuration of the search expression presentation system according to the second embodiment is the same as that shown in FIG. The correction unit 120, the learning model engine 143, and the control unit 150 shown in FIG. 1 will be described in detail.

FIG. 11 is a diagram showing the configuration of the search expression presentation system in this embodiment. Note that, in FIG. 11, the learning model engine 143 of the correction engine unit is extracted and shown. The correction unit 120 includes an input unit 1111, an acquisition unit 1112, and a correction processing unit 1113. The control unit 150 has a model construction unit 1151. The learning model engine 143 has a learning model group including a plurality of learning models, and the learning model group includes a first learning model 1161, a second learning model 1162, a third learning model 1163, and a fourth learning model 1163. It has a learning model 1164 and a fifth learning model 1165. Here, five learning models are shown as an example, but the present invention is not limited to this. Moreover, each learning model is only distinguished for convenience of explanation.

The learning model may be configured by a neural network as shown in FIG. 12, for example. The neural network includes an input layer 1210, an intermediate layer 1220, and an output layer 1230. Each layer is composed of multiple nodes. In FIG. 12, the input layer 1210 includes nodes 1211, 1212, 1213, the middle layer 1220 includes nodes 1221, 1222, 1223, and the output layer 1230 includes nodes 1231, 1232, 1233. .. The number of nodes included in each layer is not limited to the example shown in the figure, and is generally composed of a large number of nodes. The intermediate layer 1220 is not limited to one layer and may be composed of a plurality of layers. Weights are set to the respective nodes, and the value obtained by multiplying the input data by the weight is transmitted to the subsequent node. As the neural network, a CNN (Convolutional Neural Network) may be used, or an RNN (Recurrent Neural Network) may be used. Other methods may be used. The learning model may be constructed by SVM (Support Vector Machine) instead of the neural network. In the present embodiment, the learning model itself may be of any type. The learning model may be constructed by supervised learning, semi-supervised learning, or unsupervised learning.

The input unit 1111 inputs the information of the search expression received by the reception unit 110 from the user terminal 10 into at least one learning model in the learning model group.

In this embodiment, a plurality of learning models corresponding to the search items are prepared. For example, a learning model when "applicant/right holder (latest)" is specified as a search item, a learning model when "inventor (latest)" is specified as a search item, and "full text" as a search item A learning model etc. is prepared when is specified.

The input unit 1111 inputs the search keyword included in the search expression into the learning model corresponding to the search item included in the search expression input by the reception unit 110. This will be described using the specific example described with reference to FIG. The search item included in the search formula (first search formula) is “inventor (latest)”, and the search keyword is “Nomura Research Institute”. Since the search item is the bibliographic information of “inventor (latest)”, the input unit 1111 inputs the search keyword “Nomura Research Institute” into the first learning model 1161 constructed using the bibliographic information. Here, the first learning model 1161 is a learning model when “applicant/right holder (latest)” is designated, and a learning model when “inventor (latest)” is designated as a search item. As described above, the learning model may be further subdivided.

The acquisition unit 1112 acquires output data output from the learning model according to the input by the input unit 1111. In the above example, the output data output from the first learning model 1161 is acquired. For example, output data indicating "Nomura Research Institute, Ltd." is output from the learning model when "Applicant/Rightee (latest)" is specified, and "Inventor (latest)" is selected as a search item. Output data indicating “no applicable person” is acquired from the learning model when specified. Here, by using patent publications published so far as supervised learning data, what terms tend to be applicants and what terms tend to be inventors It means that the learning model that reflects is constructed. In the case of the previous example, since there is no case where the search item is "inventor (latest)" in the term "Nomura Research Institute", there is "no applicable person" and the evaluation value linked in the learning model is The high search item is “Applicant/Rightholder (latest)” and the value “Nomura Research Institute, Ltd.” is output.

The correction processing unit 1113 corrects the search formula based on the output data acquired by the acquisition unit 1112. In the above example, the search keyword is corrected to "Nomura Research Institute, Ltd." In the example of FIG. 5 described above, the search item is modified by the expert system engine 141, but as described above, it is modified based on the output data output from the first learning model 1161. May be done.

The corrected search formula (corrected second search formula) is output to the user terminal 10 by the output unit 130.

In this way, by preparing the learning model (first learning model 1161) corresponding to the search item, the output data in which the specification error of the search item or the input error of the search item is corrected is obtained from the first learning model 1161. Is output. For example, the following input errors are corrected.
-When "Judge/Examiner" is specified as the search item, enter the company name as the search keyword.-When "Agent" is specified as the search item, enter the company name as the search keyword.-"Applicant ( If you specify "Exact match)" as the search item, enter part of the company name (for example, enter Nomura Research Institute, Ltd. as "Nomura").
・When "Applicant (backward match)" is specified as the search item, enter only the beginning of the company name (for example, enter Nomura Research Institute, Ltd. as "Nomura")
・When "Judge/Examiner" is specified as a search item, enter the name of a foreigner. ・When "Applicant/Rightholder" is specified as a search item, enter the name of an English company (for example, IBM Is used in Katakana notation called International Business Machines)
・Variation in notation (for example, enter "Nomura" as "Satoyoki")
The above example is merely an example, and the present invention is not limited to this.

The second learning model 1162 is a learning model constructed using at least the classification included in the publication. Some of the classifications are used closely together. The second learning model 1162 is a learning model that has learned such classification relevance. For example, when the input unit 1111 inputs the classification code included in the search formula into the second learning model 1162, the related classification code is output from the second learning model 1162. Further, for example, when a search keyword used in full-text search or the like is input to the second learning model 1162, the related classification code is output from the second learning model 1162. The correction processing unit 1113 corrects the search formula using the related classification code, and the output unit 130 presents the corrected search formula to the user terminal 10. The output unit 130 may present a related classification code and output a notification message such as “There are N similar publications in another classification code Y” to the user terminal 10.

A learning model specific to the user is prepared in the learning model. The third learning model 1163 is a learning model constructed by using a bulletin selected by the user in advance. That is, it is a user-specific learning model. For example, the learning model engine 143 performs morphological analysis from data such as the full text of the gazette, claims, or summary previously selected by the user, and extracts a learning keyword for learning. Then, the third learning model 1163 is constructed using the extracted learning keyword and its publication as teacher data. When a search keyword is input to the third learning model 1163 constructed in this way, the related learning keyword is output as output data.

Note that, as described above, the third learning model 1163 can also be configured as a further subdivided learning model. For example, a set of learning models at the subclass level of patent classification may be generically called a third learning model 1163. Further, the publications used when constructing the learning model may include, in addition to publications previously selected by the user, cited publications or cited publications of the publications. That is, the publications described in the prior art documents in the specification of publication X, the publications cited in the notification of reasons for refusal in the application process of publication X, and the other publications cited in notification of the reasons for refusal. And other publications in which the publication X is described in the prior art document may be included.

Further, the third learning model 1163 may be constructed in consideration of the distribution status of learning keyword publications. For example, when the learning keyword W1 is distributed over the entire publication, or when the learning keyword W1 is described in the claims or the abstract of the publication, the learning keyword W1 is treated as an important keyword. You can For example, the learning keyword W1 may be treated as an important keyword rather than the learning keyword W2 that exists only in a part of the publication and is not described in an important part such as a claim or a summary. In this case, the third learning model 1163 may be used by being used as the teacher data in which the weight of the learning keyword is changed.

In addition, as the learning keyword, for example, a keyword obtained by labeling a predetermined term while the user is selecting publications may be used. Examples of labeling include “add to learning model”, “confirm later”, and “delete from learning model”. When “delete from learning model” is labeled, the learning model engine 143 deletes from the learning model by causing the corresponding learning model to perform supervised learning that reduces the feature amount of the term.

Some learning models are constructed based on search formula data that the user has input in the past. The fourth learning model 1164 is a learning model constructed for each predetermined user. The predetermined user unit may be a single user unit or a plurality of user units. For example, the user of the intellectual property department may be used as a unit, or the examiner in a specific technical field may be used as a unit. For example, in the case of an examiner, the learning model may be constructed using the search formula used in the examination. The accepting unit 110 may accept the designation of which user's mode to use, and the submitting unit 1111 submits the search formula information to the fourth learning model 1164 corresponding to the user unit for which the designation is accepted. As a result, output data corresponding to a predetermined user unit can be obtained. As a result, the search formula corrected by the correction processing unit 1113 may change for each predetermined user unit.

Further, the learning model includes a learning model constructed according to the search purpose. The fifth learning model 1165 is a learning model constructed according to the search purpose. The search purpose may be, for example, a preliminary investigation conducted by an examiner, a preliminary investigation conducted by a skilled person, an invalidation investigation, or a technical trend investigation. The fifth learning model 1165 may include a subdivided learning model constructed according to these types of search purposes. The receiving unit 110 receives the designation of the search purpose from the user terminal 10, and the input unit 1111 inputs the information of the search formula into the fifth learning model 1165 corresponding to the specified search purpose. For example, with respect to a user who has designated an invalidity investigation as a search purpose, the fifth learning model 1165 is referred to, and it is better not to narrow down the field or search for publications and utility models. The advice “should be the target” may be presented to the user terminal 10. Note that the fifth learning model 1165 may include a learning model corresponding to each country or may include a learning model for each technical field.

In this way, the search formula corrected by the correction processing unit 1113 (generated second search formula) based on the output data output from the learning model is output to the user terminal 10 by the output unit 130. The aspect in which the second search formula is presented on the user terminal 10 is as described in the first embodiment. The user terminal 10 may give an operation instruction to the presented second search formula. For example, this is the case where the menu 364 of FIG. 3 is selected on the user terminal 10. When the reception unit 110 receives this operation instruction, the model construction unit 1151 reconstructs the learning model by using the items related to the correction portion designated by the operation instruction as the teacher data. The learning model to be reconstructed may be reconstructed only the learning model used for the correction of the correction part, or may be reconstructed including the learning model related to the learning model, All learning models in the learning model group included in the learning model engine may be reconstructed.

In the example described above, five types of the first learning model 1161 to the fifth learning model 1165 are described for convenience of description, but the present invention is not limited to this. Each learning model may be further subdivided, or a learning model in which a plurality of learning models are integrated may be used. The input unit 1111 may input the information of the search formula into at least one or more learning models, or may input the information of the search formula into a plurality of learning models.

FIG. 13 is a diagram showing an example of a search formula presentation method in the search formula presentation system according to the present embodiment.

In step S1301, the reception unit 110 receives an input of a search expression from the user terminal 10.

In step S1302, the correction unit 120 identifies a search item included in the search formula.

In step S1303, the input unit 1111 inputs search formula information into the learning model corresponding to the search item specified in step S1302. Both the search item and the search keyword may be input, or the search keyword may be input.

In step S1304, the acquisition unit 1112 acquires the output data output from the learning model in response to the input in step S1303.

In step S1305, the correction processing unit 1113 corrects the search expression input in step S1301 based on the output data acquired in step S1304.

In step S1306, the output unit 130 outputs the search formula modified in step S1305 to the user terminal 10.

As described above, according to this embodiment, the information of the search formula is input to the learning model, and the search formula is modified based on the output data output from the learning model. Therefore, the correction is not limited to the uniform (or mechanical) correction performed by the expert system engine 141 or the thesaurus dictionary engine 142, but the correction closer to the correction performed by a human such as an expert is performed. Therefore, an appropriate search formula can be presented to the user terminal 10. Further, even if the input search formula has no problem in form, it is possible to output to the user terminal 10 a search formula as if it was modified by a user who is familiar with the search practice.

In the above embodiment, the information of the search formula is input to the learning model and the search formula is corrected based on the output data output from the learning model. However, in addition to these correction candidates, the past Among the executed search formulas, those with high similarity/relevance are sorted and displayed as a reference, and elements of the search formula (part of the search formula) are cut out as correction candidates from the displayed reference search formula. May be possible. Here, since the past search formula is input in the process of constructing the learning model, it is possible to take out the past search formula similar to the search formula currently being corrected. Also, in addition to the fact that the search formula is modified based on the output data output from the learning model, the configuration that displays the past search formula with high similarity/relevance was explained. Alternatively, the past search formula having a high similarity/relevance may be displayed without modifying the search formula based on the above.

Each unit for realizing the functions of the above-described embodiments can be implemented by, for example, hardware or software. When implemented by software, program code for controlling hardware may be executed by various processors such as a CPU and MPU. Hardware such as a circuit for realizing the function of the program code may be provided. Part of the program code may be implemented by hardware, and the rest may be executed by various processors.

The invention described in the above-described embodiment can be expressed as the invention described in the scope of claims, and can also be expressed as the invention described in the following additional notes. Further, the present invention is also a combination of the contents described in the claims and the contents of the following additional description.
[Appendix 1]
A reception unit that receives a search expression input from the user terminal,
A correction unit that corrects the search expression that has received an input based on a learning model,
A retrieval formula presentation system having an output unit for outputting the corrected retrieval formula to the user terminal.
[Appendix 2]
Of a learning model group including a plurality of types of learning models, a predetermined learning model, the input unit that inputs at least a part of the search formula,
An acquisition unit for acquiring output data output from the learning model according to the input by the input unit;
The search formula presentation system according to appendix 1, further comprising a correction processing unit that corrects the search formula based on the output data.
[Appendix 3]
The input unit, among the learning model group including a plurality of types of learning models corresponding to the search item, a search included in the search formula in a learning model corresponding to the search item included in the search formula that receives an input. The search expression presentation system according to attachment 2, which inputs a keyword.
[Appendix 4]
The search formula presentation system according to Appendix 3, wherein the search items include items relating to bibliographic information, and the learning model group includes a first learning model constructed using at least the bibliographic information.
[Appendix 5]
The search expression presentation system according to appendix 3 or 4, wherein the search items include items relating to classification, and the learning model group includes a second learning model constructed using at least the classification.
[Appendix 6]
The search item includes an item relating to at least one of the full text of the publication, the scope of claims, and the abstract, and the learning model group uses at least the full text of the publication, the scope of claims, or the abstract of the publication selected in advance. The search expression presentation system according to any one of appendices 3 to 5, including the constructed third learning model.
[Appendix 7]
The search formula presentation system according to Appendix 6, wherein the third learning model is constructed using learning keywords obtained from the preselected publications.
[Appendix 8]
The learning keyword is obtained from the preselected publication by morphological analysis, is obtained from the preselected publication according to a user's instruction, or is distributed according to the distribution appearing in the preselected publication. The search expression presentation system according to appendix 7, which is obtained by at least one of the obtained methods.
[Appendix 9]
The learning model includes a fourth learning model constructed for each predetermined user,
The reception unit receives the designation for each user,
9. The search formula presentation system according to any one of appendices 2 to 8, wherein the input unit inputs at least a part of the search formula into the fourth learning model constructed in units of the designated user.
[Appendix 10]
The learning model includes a fifth learning model constructed according to a search purpose,
The reception unit receives the designation of the search purpose,
10. The search formula presentation system according to any one of appendices 2 to 9, wherein the input unit inputs at least a part of the search formula into the fifth learning model constructed for the designated search purpose.
[Appendix 11]
The reception unit further receives an input of the number of target search results from the user terminal,
The correction unit generates a search formula that is a number of cases within a predetermined range from the number of cases,
The search formula presentation system according to any one of appendices 1 to 10, wherein the output unit outputs the search formula to the user terminal.
[Appendix 12]
The reception unit further receives, from the user terminal, inputs of designation of a plurality of modes,
The search formula presentation system according to any one of appendices 1 to 11, wherein the correction unit generates a search formula according to the mode.
[Appendix 13]
A step of receiving a search expression input from the user terminal,
Modifying the search expression that received the input based on a learning model,
Outputting the modified search formula to the user terminal.
[Appendix 14]
A program for causing a computer to function as each unit of the search expression presentation system according to any one of appendices 1 to 12.

10 user terminal 11 input unit 12 display unit 100 search expression presentation system 110 reception unit 120 correction unit 130 output unit 140 correction engine unit 141 expert system engine 142 synonym dictionary engine 143 learning model engine 150 control unit 160 search processing unit

Claims (14)

A reception unit that receives the first search expression input in the user terminal,
A correction unit that corrects the first search expression that has received the input and generates a second search expression using a plurality of correction engines ;
As modifications were made locations are identifiably displayed on the user interface screen of the user terminal, identification of the the second search formula, modification engine that conducted the type or modification of modification engine that fixes An output unit that outputs information and the information to the user terminal. A reception unit that receives the first search expression and the designation in user units input in the user terminal,
At least using a learning model, a correction unit that corrects the first search expression that has received the input to generate a second search expression,
An output unit for outputting the second search formula to the user terminal so that the modified portion is displayed in an identifiable manner on the user interface screen of the user terminal;
Have
The learning model is a learning model constructed for each predetermined user, and the information for each designated user is input.
Search formula presentation system. The output unit outputs the second search expression to the user terminal so that the second search expression is displayed on the user interface screen together with the first search expression that has received the input. The search formula presentation system described in 1 or 2 . The search expression presentation system according to claim 1 , wherein the correction engine includes at least one of an expert system engine, a thesaurus dictionary engine, and a learning model engine. The reception unit can further receive an operation instruction from the user terminal for a portion where a correction is made in the second search formula,
The search-type presentation system according to claim 4, further comprising a control unit that performs control according to the operation instruction. The control unit reconstructs a learning model used by the learning model engine based on the corrected content when the operation instruction indicates that the content of the correction is to be learned. Search presentation system. The search formula presentation system according to claim 1, wherein the output unit outputs the user interface screen to the user terminal. The search formula presentation system according to any one of claims 1 to 7, wherein the output unit outputs recommended search conditions that are not included in the second search formula to the user terminal. The reception unit further receives an input of the number of target search results from the user terminal,
The correction unit generates a third search expression that is a number of cases within a predetermined range from the number of cases,
The search formula presentation system according to any one of claims 1 to 8, wherein the output unit outputs the third search formula to the user terminal. The reception unit further receives, from the user terminal, inputs for designating a plurality of modes,
The search formula presentation system according to any one of claims 1 to 9, wherein the correction unit generates the second search formula according to the mode. The correction unit corrects the first search formula using at least a learning model,
The reception unit receives designation of a search purpose,
11. The learning model is a learning model constructed according to a search purpose, and at least a part of the first search formula is input for the specified search purpose. The search formula presentation system described in. By a computer communicatively connected to the user terminal,
A step of accepting a first search formula inputted at the user terminal,
Modifying the first search expression that has received the input to generate a second search expression using a plurality of modification engines ;
The second search formula and the type of the correction engine that has been modified or the identification information of the modification engine that has been modified so that the location where the modification has been performed is displayed in an identifiable manner on the user interface screen of the user terminal. And a step of outputting to the user terminal. By a computer communicatively connected to the user terminal,
A step of accepting a first search expression and a user-specific designation input at the user terminal;
Using at least a learning model, modifying the first search expression that has received the input to generate a second search expression;
Outputting the second search formula to the user terminal so that the modified portion is displayed in an identifiable manner on the user interface screen of the user terminal.
Have
The learning model is a learning model constructed for each predetermined user, and the information for each designated user is input.
Search formula presentation method. The computer program to function as each unit of the search expression presentation system according to any one of claims 1 to 11.