JP2020016964A - Case component learning program - Google Patents

Abstract

To cause a computer to learn a case component on the assumption that scopes of patent claims are automatically generated.SOLUTION: A computer is caused to execute a learning step of learning verbs extracted from character strings and case components modifying them as data sets one after another and then generating a learnt model having relation between the verbs and case components defined as association levels of three or more stages. At this time, there may be further included: a character string input reception step of receiving input of a character string; and a search step of referring to the learnt model generated in the learning step, and then searching for, when the character string input in the character string input reception step is a verb, a case component based upon the association level of the input verb and searching for, when the character string input in the character string input reception step is a case component, a verb based upon an association level of the input case component.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a case component learning program for extracting and learning a case component from a character string described in an electronic data document such as a patent specification.

  Conventionally, a case component extraction program for extracting a case component from a character string defined in the claims of the patent specification has been proposed.

JP 2014-048987 A

  However, in the technology disclosed in Patent Literature 1 described above, even though the case component can be automatically extracted, it is particularly disclosed that the claims are automatically created using the actually extracted case component. It has not been.

  The present invention has been devised in view of the above-described problems, and has as its object to provide a case component learning program for learning case components on the assumption that claims are automatically created. Is to provide.

  The case component learning program according to the present invention sequentially learns a verb extracted from a character string and a case component relating to the verb as a data set, so that the relation between the verb and the case component is three or more levels. And causing the computer to execute a learning step of creating a learned model defined as

  The case component learning program according to the present invention learns a phrase structure having a verb and one or more case components related to the verb and a proposition corresponding to the verb as a data set sequentially, thereby obtaining the phrase structure and the proposition. And causing the computer to execute a learning step of creating a learned model whose relationship with is defined as three or more degrees of association.

  The case component learning program according to the present invention sequentially learns a dependency structure of a verb and one case component related to the verb as a data set, and obtains the dependency structure and the morpheme. And causing the computer to execute a learning step of creating a learned model whose relationship with is defined as three or more degrees of association.

  It is possible to narrow down the candidates of the verb to which the case component depends in the claims, and to improve the convenience even if the user is prompted to confirm the last part.

It is a figure for explaining a case component. It is another figure for demonstrating a case component. FIG. 9 is a diagram showing an example in which a verb extracted from a character string and case components related thereto are sequentially learned as a data set. FIG. 9 is a diagram showing an example in which a phrase structure having a verb and one or more case components related thereto and a proposition corresponding thereto are sequentially learned as a data set. FIG. 9 is a diagram illustrating an example of a dependency structure of a verb and one case component related to the verb, and a morpheme related to the dependency, which is sequentially learned as a data set.

  Hereinafter, as a mode for carrying out the present invention, a case component learning program for extracting and learning case components from a character string will be described in detail with reference to the drawings.

Theoretical explanation of the present invention

  A case component has been proposed as an idea for quantifying and quantifying the breadth of the technical range of the patented invention.

  The claims are described in a so-called requirement enumeration system in which each of the constituent elements A, B, and C is defined by enumerating them as "a device (method) including ~ A, ~ B, and C". , Each component A, B, C is associated with a verb phrase. That is, when the constituent elements A, B, and C are the subjects, they are each associated with a verb phrase, and a sentence composed of the subject and the predicate can be formed. When forming a sentence as a predicate, a verb in these verb phrases basically determines what combination of noun phrases is used in order to realize the movement, state, and relationship expressed by the verb. If a verb tentatively calls the combination of noun phrases necessary to form a sentence according to the lexical meaning of the vocabulary that the verb possesses, it is a verb, The noun phrase required as a sentence component can be said to be a case component in supplementing the verb. Further, the components A, B, and C are also related to the verb “provided” related to the ultimate component “XX device”. Therefore, according to the present invention, the components A, B, and C may also be counted as case components. That is, the present invention may extract a case component as a noun phrase related to all the verbs included in the character string constituting the claims.

  For example, it is assumed that the component "signal generating means" is defined by the following B-1).

  B-1) "Signal generation means for generating a drive signal in response to a request from a user, ..."

  At this time, a verb phrase "generate a drive signal in response to a request from the user" is received by the signal generation means. In order to realize and complete the action of the verb “generate” in this verb phrase, the function of requesting a noun phrase “according to a request from a user (according to)” and “drive signal (a)” is dominant. Yes, these noun phrases are case components. Noun phrases such as "request (from user)" and "drive signal (") "are used in relation to the verb" generate ", respectively. The meaning is case. The lexical meaning of the noun phrase that realizes this case component and the case formed by the relational meaning of those noun phrases should be realized by this verb “generate” At the core of the proposition. In the above example, the operation start condition for realizing the proposition with the verb "generate" must be "according to the request from the user", and the "drive signal" Therefore, these two are the condition numbers for realizing the proposition by the verb.

  FIG. 1 schematically illustrates a case where the number of conditions that need to be cleared for realizing a proposition using such a verb is small. Since the number of conditions for starting an action by a verb is reduced, the possibility of realizing a proposition is improved. On the other hand, FIG. 2 schematically shows a case where the number of conditions that must be cleared for realizing a proposition by a verb increases. In such a case, the possibility of realizing the proposition decreases. .

  As the number of conditions increases, the number of steps for determining whether or not the conditions are satisfied before the operation is actually started increases. Even considering that there is some difference in the probability of satisfying each condition, the verb phrase is affected as the number of judgment steps of this condition increases, in other words, as the case component that defines the condition increases. This means that the possibility of falling under the component A is reduced, indicating that the technical scope is reduced by the reduced possibility. Conversely, when the number of case components is small, it means that the possibility that the verb phrase corresponds to the component A to be affected is increased, which indicates that the technical range is expanded accordingly. .

  In this way, the number of case components (condition number) governs the possibility of starting an action by a verb, and furthermore the possibility of realizing a proposition, and this affects the breadth and width of the technical range. By counting the number of case components included in this verb phrase using the minimum extraction unit of the binning method as the case component, it is considered that the digitization according to the extent of the technical range can be realized.

  In the example of B-1) described above, the two case components “in response to a request from the user” and “the drive signal are transmitted in response to the verb“ generate ”for realizing the proposition of the component“ signal generation means ” ”, The number of case components is 2. At this time, if there is no case component “in response to a request from the user” that defines the operation start condition of the verb “generate”, the number of case components is 1, regardless of whether or not there is a request from the user. Therefore, the "drive signal" can be "generated" at any time, and the possibility of realizing the proposition of "signal generation means" is increased. This suggests that the component "signal generation means" can improve the possibility that the identity of the technical element of the product infringed can be proved. It is considered that this corresponds to the easiness of capturing the patented invention on the product, and thus the technical scope of the patented invention. For this reason, by expressing the proposition feasibility of the components through the number of case components, it is considered that more optimal numerical value and quantification can be achieved according to the extent of the technical range.

  When actually determining the number of case components of each component, in the example of B-1) described above, the case component "in response to a request from the user" "the drive signal" However, the actual extraction of this case component is performed using morphemes such as "accordingly" and "".

  Table 1 shows an example of a morpheme that is referred to as a mark when extracting a case component.

  The claims are defined not only in the orthodox form such as B-1), but also in the case defined in, for example, C-1).

  C-1) "Reflecting means for totally reflecting the light beam whose spot diameter is controlled by the first lens by the reflecting plate, ..."

  For example, the verb "total reflection" is applied to the component "reflecting means" constituting the claims. There are "luminous flux (") "and" reflector (")" as forms that define the operation start condition of the verb "to be totally reflected." Can be. However, when focusing on one of the case components “luminous flux”, it can be seen that the verb phrase “the spot diameter was controlled by the first lens” is further affected. Rather than defining all luminous fluxes as "totally reflect" objects, defining the luminous flux to which conditions are added as "totally reflect" objects is technically limited and adds a component "reflecting means". Proposition feasibility decreases. For this reason, it is necessary to correct the decrease in the possibility of proposition realization as the number of case components.

  In such a case, in the verb phrase “the spot diameter is controlled by the first lens” related to “luminous flux”, two case components “first lens (first lens ( ) "And" spot diameter () "are present, so the case component number 2 is further added. As a result, the total number of case components in C-1) is four.

  The description C-1) of the claims may be defined as C-2) below depending on the creator of the specification.

  C-2) “Light flux control means for controlling the spot diameter of the light flux by the first lens, and reflection means for totally reflecting the light flux by the reflector”

  The technical range formed by the "light flux controlling means" and the "reflecting means" in C-2) is substantially the same as C-1). In C-2), the content of the verb phrase related to "luminous flux" in C-1) is redefined as a proposition to be realized by "luminous flux controlling means". Actually, in C-1), some means or members are used to create the defined "luminous flux", but only the "luminous flux controlling means" is not defined. . Similarly, when the number of case components is also counted for C-2), the number of case components is 2 by the light flux control means and the number of case components is 2 by the reflection means, so that the total number of case components is 4. Become.

  As described above, since the breadth of the technical range is counted through the number of case components required to realize the propositions of the components, the description method and the number of words are substantially different from each other. C-1) and C-2) having the same technical range can be expressed by the same number of case components. If the minimum extraction unit is a component, C-1) is 1 and C-2) is 2. If the minimum extraction unit is a word, C-1) is 7, and C-2) is 9. In consideration of this, it is considered that this number of case components can be a minimum extraction unit of a more suitable numerical method.

The configuration of C-1) and C-2) may be defined by the following C-3).
C-3) “Light flux controlling means for controlling the spot diameter of the light beam by the first lens, and reflecting means for totally reflecting the light beam whose spot diameter is controlled by the light beam controlling means by the reflector,”

  The underlined portion of C-3) corresponds to the "such light beam" portion in C-2). However, the content of the verb phrase described in this wavy line is a proposition that has already been realized by the straight line portion related to the "luminous flux control means", and the description of the wavy line part is "reflection". This is merely a confirmation to prevent the interpretation of the "luminous flux" as an object to be totally reflected by the "means". For this reason, when calculating the number of case components, if the description of the wavy portion is also counted in the same manner, the contents of the underlined portion are double-counted, resulting in deterioration of measurement accuracy.

  For this reason, by paying attention not to count again the description of the dashed line portion that has already been executed as a proposition, it is possible to eliminate the difference in the number of case components due to the difference in the description order and the description method of the claims. Actually, it is necessary to determine whether or not the proposition realized by the description of the straight line portion and the description of the wavy line portion are the same.

  The numerical method using the minimum extraction unit as a case component is particularly useful when judging the identity of a proposition. Case components, unlike words, have the property of being governed by a verb, so that the same case component means that in addition to the content of the case component, the verb that governs it must also be the same. I have. If both the case component and the verb that governs it are the same, the proposition realized by this is also the same. Therefore, it is possible to determine whether or not the propositions are the same through the determination of the identity of the case components, and thus it is possible to easily determine whether or not the proposition is a double count. Further, the difference in the description method between the creators does not affect the obtained numerical value.

  Note that, in the above-described example, the case where the description of the claims is described in the form of constituent requirement enumeration has been described as an example. However, even if the description is described in another format, it can be similarly counted. For example, in the case where the information is described in Jepson type, the description of “Kari-Written”, which is a known part, is counted by the same method. At this time, if the counting method described in “write” is not known, the counting may be performed after rewriting from the Jepson type to the component requirement enumeration type. Incidentally, the method of rewriting from the Jepson type to the constituent requirement enumeration type has been conventionally known.

  Further, even when the claims are described in a newly-written type, they can be counted by the same method. For example, if C-2) is described in a new type, the following C-4) is obtained.

  C-4) "The spot diameter of the light beam is controlled by the first lens, and the light beam is totally reflected by the reflector, ~"

  C-4) described in this rewritten type is different from C-2) in that the terms "luminous flux controlling means" and "reflecting means" are omitted, but otherwise is the same. . The "flux control means" and the "reflection means" are subjects and are not counted, and count only the noun phrases related to this subject. The number of components does not differ.

Second embodiment

  When extracting a case component, it may be based on the following rules. In the second embodiment, the portion underlined corresponds to one case component to be extracted. Hereinafter, what is referred to as "count" means that each is extracted as a case component.

(1) Cases in which words that include possibility or flexibility are added to nouns

  Words described in the verb meaning "noun + can", such as noun + possible, noun + possible, noun + freely, noun + free, etc., are included in case components to be extracted. Absent

  Example) Can be placed → Noun + ~ Can be counted because it is a verb meaning that it can be done.

Not counted because it is an adjective meaning that detectable → noun + possible.
"Synchronizable"-> Can not be counted.
Input possible, protruding / storing possible, selectable → Not possible to count.
The height position can be changed, but not counted.
Free to slide → Free, and possible, do not count.

  That is, as in the case where "-possible", "-free", "-free", etc. are added to a noun, words including the possibility or flexibility of the noun are added to the extracted character string. If the extracted character string is included, an algorithm that excludes the character string from the case components to be extracted may be provided.

(2) Subject rules

  1) Extract the subject first. All the items that are "●●●,", "●●● is," "●●● is," "●●● also," etc. are regarded as the subject and all are extracted. (At this time, ●●● has nothing to do with “above”, “pertinent”, “above”, etc.)

  2) Sort and check whether the extracted subject (●●●) is described earlier. As a result, if there is a subject that matches 100% of the extracted subject (●●●) or contains a part of the subject, the process proceeds to 3). In other cases, since the extracted subject (●●●) appears for the first time, the extracted subject (●●●) is specified as a case component.

  3) When shifting to 3) in 2) above, none of them are counted as case components.

  That is, it is first determined whether or not the relationship between the subject and the verb is included in the extracted character string. As a result, when the relationship between the subject and the verb is included in the extracted character string, it is confirmed whether or not the text corresponding to the subject has already been described in the upper part. If the word corresponding to the subject is already described in the upper part, the word corresponding to the subject is excluded from the case components to be extracted, and the word corresponding to the subject is still described in the upper part. If not, since it is newly appeared, the word corresponding to the subject is included in the case component to be extracted.

  This makes it possible to eliminate the irrationality that the case component increases as the number of subjects increases in the description of the claims having many subjects.

  A program that performs the above process may be provided.

(3) Double definition rule

(Explanation of double definition logic)
First, the subject to which the verb is related is specified, and then a case component that satisfies the operating condition of the verb is extracted.

For example, in the case of the following sentence, the “subject that the verb depends on”, the “verb”, and the “case component” are specified as follows.
"Rotation angle signal detecting means for detecting a rotation angle signal output from the output shaft of the internal combustion engine"

  Such a list is created sequentially. It is not essential to create a list, and a pair of “subject subject to verb”, “verb”, and “case component” may be stored.

  In the subsequent stage, for example, it is assumed that the phrase “based on the rotation angle signal detected by the rotation angle signal detection unit” is defined.

  The following logic determines whether or not the "rotation angle signal detection means", which is related to the verb "detected", is counted as a case component.

  The pair of "the rotation angle signal detecting means" and "detected (performed)" matches the previously stored one. Therefore, it is not counted because this is a proposition that has already been executed.

Since "~ detected rotation angle signal " is not related to the verb "detected" (conversely, it is modified by the verb), the case component is related to the verb "detected" in the first place. It is not necessary to judge whether or not this is the case component.

  In this way, a pair of “subject subject to verb”, “verb”, and “case component” is stored, and these “subject subject to verb”, “case component” for “verb” are stored. In the case where the component “related” appears later, it is not counted as a double definition.

Example 1 Claim 1 of p2000-196550
In a vehicle equipped with an internal combustion engine as a drive source, a vehicle speed detection unit that detects a vehicle speed of the vehicle; a braking operation detection unit that detects a state of an operation input to a braking device of the vehicle operated by a driver; Exhaust gas recirculation means for recirculating a part of the exhaust gas from the exhaust path into the intake pipe, and when it is recognized that the vehicle is in a stopped state based on the vehicle speed detected by the vehicle speed detection means, the exhaust gas recirculation means enters the intake pipe. Control means for recirculating the exhaust gas and reducing the amount of exhaust gas recirculated by the exhaust gas recirculation means when the braking operation detection means detects release of the operation input in the stopped state. Control device for internal combustion engine for vehicles. "

  In this case, the following list is created.

For example, 6 based on “based on the vehicle speed detected by the vehicle speed detection means” is
"By the vehicle speed detecting means" is not counted as a case component because it has already been executed at 2 in relation to the verb "detected".

  The phrase "recirculate exhaust gas into the intake pipe by the exhaust gas recirculation means" of 8 is related to the verbs "recirculate" by "the exhaust gas recirculation means" and "recirculate" in the intake pipe. "Exhaust gas", which is not counted, is counted as one major component because it first appeared because of the verb "recirculate".

  As described above, it is not necessary to focus only on the pair of “verb” and “subject to which the verb is dependent” to check whether the execution has already been performed. Also pay attention.

  Regarding 9 “release of the operation input is detected by the braking operation detecting means”, “detected” by “braking operation detecting means” and “detected” are not counted because the condition of 4 is satisfied. The cancellation of operation input appears for the first time. (The cancellation of the operation input) is counted as a case component.

  The point is that when referring to the list in the course of this extraction, it must be taken into account that the active and passive voices can be changed for the verbs described in the list.

Example 2 Claim 1 of P2000-258811
"In a printer control device for controlling a plurality of types of printers each having a different resolution, an image selecting means for selecting an image to be printed from a plurality of images, the number of pixels of the image selected by the image selecting means, A printer deciding unit for deciding a printer to be used from the plurality of types of printers based on the resolution, and image data representing an image selected by the image selecting unit to the printer decided by the printer deciding unit. A printer control device including a transmission unit for transmitting a printer. "

  In this case, the following list is created.

  The "number of pixels of the image selected by the image selecting means" of 4 has already been executed because there are 3 pairs of "the image selecting means" for the verb "select".

  Similarly, the "6 to the printer determined by the printer determining means" has already been executed because there is a pair "5 by the printer determining means" for the verb "determine".

  The “image selected by the image selecting unit” of 7 has already been executed because there are 3 pairs of the “image selecting unit” for the verb “select”.

  Those already executed are not counted as case components.

  That is, in the rule of the double definition, it is ultimately determined whether to include or exclude the case component to be extracted based on the following algorithm.

Step 1

  Based on the verb, the case component involved in the verb and the subject involved in the verb are specified. Next, it is checked whether the same verb is present in the upper row (even though there is a difference between active and passive). As a result, if there is no identical verb in the upper row, the double definition check is not performed. On the other hand, if the same verb is present in the upper row, the process proceeds to step 2 below.

Step 2

  For each verb, the identity of the subject and the case component to which the verb depends is determined. The above is the difference between the "subject subject to the verb" and the "case component" for the same verb (No. 5) that appeared first, and the case component for the same verb (No. 8) that comes out later. This is an example of determining the identity.

Step 3

  If there is an identical case between the "subject subject to the verb" and the "case component" for the same verb that appeared first and the case component for the same verb that comes later, Does not regard the case component of the same verb that comes out of as a case component (because the proposition has already been executed in the upper part).

  The above processing operation is similarly performed for all other verbs.

  That is, as an algorithm, verbs are sequentially picked up. Check if the verb you picked up has already appeared before. If they have appeared, the identity between the “subject subject to the verb” and the “case component” for the same verb that first appeared and the case component for the same verb that came out later is determined. As a result of the determination, if the same verb comes out, the case component of the same verb that comes out later is not regarded as the case component.

  At this time, it may be realized through the following steps.

  An identification step of sequentially extracting verbs and identifying whether the extracted verbs are already defined in the upper part thereof; As a result of the identification in the identification step, if it is identified that the extracted verb has already been defined in the upper part, the noun or noun phrase related to the verb already defined in the upper part, and the upper part A specifying step of specifying a noun or a noun phrase indicating the subject to which the verb defined in step 1 is related. An identity determination step of determining identity between the noun or noun phrase identified in the identification step and the noun or noun phrase related to the extracted verb. An exclusion processing step of excluding, from the case components to be extracted, those nouns or noun phrases that are determined to be the same among the nouns or noun phrases related to the extracted verb.

  In the identification step, words composed of a plurality of nouns or noun phrases connected by morphemes may be sequentially extracted, and identification may be performed to determine whether the extracted words have already been defined in the upper part. Then, in the exclusion processing step, as a result of the identification in the identification step, when it is identified that the extracted word has already been defined in the upper part thereof, the extracted word is excluded from the case components to be extracted. It may be.

For example, in the case of " When connecting to the multi-control unit ...", the underlined part is extracted as the case component for the first time. Further, in the same claim, if the same wording appears in the second and subsequent times as " when connected to the multi-control unit ," these are combined into one in order to avoid double counting. A processing operation is performed so as to be a case component or not to extract them as a case component.

  That is, it is also checked whether "noun or noun phrase" + "morpheme" + "noun or noun phrase" is not defined twice.

(4) About morpheme

  In the case where a word consisting of “noun or noun phrase” + “A” + “noun or noun phrase” is extracted, A = “to”, “to”, “to”, “at”, “ya” , The individual “nouns or noun phrases” separated by the “A” are separately extracted as case components.

  When A = “of”, each “noun or noun phrase” separated by this “A” is extracted as one case component.

(5) The name of the invention of the preamble is not counted

  In the description of the claim "B is characterized by ... in A", "in -A" is called "preamble". In this preamble, the name of the invention is often entered for A. Regarding the title of the invention, the name of the invention is often entered for B even at the end of the claims, "B characterized by ...". Such a title of the invention is not extracted as a case component. The wording of the title of the invention may be cited from the description of the title of the invention in the specification. B often matches 100% with the title of the invention, but A often does not. For this reason, A [noun] in front of “Oto,” and “Dat,” is picked up, and if this includes a part of B (assuming that the name of the invention matches 100%), the invention May be regarded as a name and a process of not extracting this as a case component may be performed. In other words, the meaning of A is that even if it does not coincide 100% with the title of the invention, it is only necessary that it overlaps a little.

  That is, in this (5), if the extracted character string contains a word that is the same as or a part of the name of the invention, a process of excluding the word from the case components to be extracted is performed. It may be done.

(6) Pronouns For pronouns composed of any of "this", "it", "that", "which", "here", "there", "there", "where", the case to be extracted You may make it include in a component. Even if these are pronouns, they function as the case components described above, and should originally be included in the case components. This enables highly accurate case component extraction.

  On the other hand, adverbs (any of "this", "so", "oh", "how") and adverbs ("this", "that", "that", "any") are extracted You may make it exclude from the case component which should be.

(7) Regarding reference to the specification In the automatic extraction, it may be confusing whether it should be determined as one noun or whether it is composed of two or more nouns. In such a case, the description in the specification may be referred to.

  For example, there is a case where the case component is extracted separately from “back tilt range of backrest” → “back” and “back tilt range”. In some cases, the case component may be extracted by dividing the case component by “tilting the backrest → tilting the back”. There is a case where the case component is divided and extracted even though it is a single case component at the back of the backrest. First table writing means → case component may be extracted separately. Printer allocation instruction means → case component may be extracted separately.

  For this reason, a check function based on the description in the specification is provided. For example, in the case of the above-mentioned example of “the back tilt range of the backrest”, the description of the reference numeral in the specification should include the configuration of “the backrest”. By referring to the description of the reference numerals, it is possible to avoid cutting the “back” with the “back”. Also, in [Embodiment], since there is a sign in the configuration such as ●●● 5, ●●● 8, etc., it should be possible to sort by therefrom and pick out nouns to check. In other words, not only the scope of the claims but also the description in the specification is positively utilized for checking. That is, when extracting a noun constituting a case component from a character string extracted from the description of the claims, a determination is made as to whether or not the noun is a character string constituting the noun while referring to the description in the specification. I do. Thereby, the extraction accuracy of the noun can be improved.

(8) Replenishment of OR Rule When there is a phrase “at least one of”, items related to “at least one of” are counted together as one rank component.

  Example) "At least one of those bases" → 1 rank component. "The lower part of at least one of the left and right sides of the backrest." "At the rear end of the seat at the rising base end or at a position forward of the base end" → "one component at a time". "Whether they are arranged side by side in the same direction or arranged in a direction orthogonal to each other" → "or", so they are collectively one rating component. “At least one of bin users” → collectively one rating component. “Image information of input page information” → collectively one rating component.

  That is, if the extracted character string is followed by a phrase indicating a plurality of nouns or noun phrases, and if a condition indicating “at least one” of the plurality of nouns or noun phrases is added, the plurality of Nouns or noun phrases are extracted as a single case component without being individually extracted as a case component.

(9) Other

  A word displayed by “verb” + “koto” is extracted as one case component by itself. For example, “being operated” → “being operated” is a verb, but “thing” constitutes one noun phrase. Noun phrases become case components. "Elapsed" → counts because it is a noun form. That is, the noun form of the verb in which the verb is connected to "koto" is extracted as a case component.

  Also, “abbreviated”, “slightly”, and “sequentially” are excluded from case components to be extracted.

  Further, when there is a word composed of “noun” + “no” + “adjective” + “noun”, these may be collectively extracted as one case component. For example, the “GPS satellite having a large elevation angle” corresponds to this form, but these are collectively extracted as one case component.

  In addition, there is a phrase such as “numerical value to numerical value” that determines a numerical value range through “to”. For example, “Temperature 19-30 ° C.” can be rewritten as “Temperature 19 ° C. to 30 ° C.”, and “kara” is a morpheme, so 19 ° C. and 30 ° C. are extracted as independent case components, respectively. They should not be extracted together as one case component. In other words, when such a description is made, the numerical values separated by “to” are extracted as case components.

Third embodiment

  When extracting a case component, it may be based on the following rules. In the third embodiment, the portion underlined corresponds to one case component to be extracted. In the following, what is referred to as “count” is whether or not to count as a case component.

(1) The rule of "in case" The following morpheme Z (in case, when, during, between, whether or not, every time, every time, every time, at the time, If a clause containing any of the following is detected: A, B, C is a noun or a noun clause,
If “A is B as (verb) + Z” and A is already defined above the description, the description is collectively extracted as one component, and A is If not defined in the upper part, A and B are extracted as separate case components.

(1-1) When A has already appeared in the upper row
If A outputs B → one component at a time.
When A for C outputs B → Only B + A is counted as one case component, and C is counted separately.
The reason is to follow the rule of the following subject (2).
For example, when the total data amount exceeds the free space of the recording medium → 1 number of components. "The total data amount" has already been issued, so no count.

(1-2) When A is in the upper row and has not yet appeared
When A outputs B → 2 case components.
When A for C outputs B → A, B, C are counted separately.
Because it follows the rule of the subject of (2).
For example, when the size of the seed cell is equal to or smaller than the threshold value → the number of components per case. Since "seed cell size" was first introduced, it was counted as a case component.

(1-3) Other example When Δ △ (verb) is used. ・ When B is added to A → 2 case components.
・ If B is △△ → 1 rank component.
・ If A is → → 1 rank component.
・ In the case of A, → 1 rank component.
The number of components per rank, regardless of whether A appears in the previous stage or not. For example, when the abnormality is detected → Number of components per case

(1-4) Another case where the rule of (1) is applied is, for example, the following morpheme.
“When” = when, during, during, between, or not, each time, with (when input = input), at every time, every time,ほ ど, 中, 状態, 状態, 〜, 〜, 態 様, 態 様 (whether or not), and も (in some cases).
In addition, the same applies to the case where "

(1-5) Example of (1-4)
When canceling the processing result involving the addition or deletion of the data pointer , = when the processing result involving the addition or deletion of the data pointer is canceled . That is, it is a second case component according to the rules of the case.

(2) Subject Rule (2-1) Principle Rule ●● is not counted even if ●● is not a constituent element ... (A).
OO is ... OO is the first thing to come out and counts ... (B).
At this time, even in the case of (A), ●● is ... (active) → Do not count. Because it is just a limitation of the configuration already defined.
●● is ~ (passive) → count. This is because the operator agrees to perform ●, and the processing operation is executed once. In other words, coordinate counts separately the processing state → coordinate with the processing state of the processing relating to entering the processing state = coordinates of a process related to being input. Number of two components.

For example, the processing state of the processing relating to the input of the coordinates → the coordinates are the first to appear (pattern (B)).
For example, the character conversion is performed on the character data . However, since the verb is in the passive form of "performed", "code conversion" is counted as a case component.
For example, when the exhaust valve opens and closes → pattern (A) because the exhaust valve has come out before. Since the verb “open / close” is active, the “exhaust valve” is not counted as a case component.
For example, when the first error is detected → According to the rule of the subject, since “detected” is a passive form, it is a good idea to call it the number of one case component. Number of components per case.

(2-2) Even if the subject has a plurality of cases, the rule described above is followed.
For example, if A and B are C → A and B come out newly, three case components, and if they already appear, one case component (A and B together If you think that it is C, you have 1 case component). For example, A and B are set to C-> Since they are passive, they are counted individually. Three case components.

(3) Rule of constituent elements (3-1) Pattern described in claims and number of case components (3-1-1) Pattern of product invention Defined by completely writing out inventions with exactly the same invention and defining requirements Then, it becomes the following.
・ Writing method " ~ B is arranged in ~ A ,
C consisting of this B are connected,
An apparatus characterized in that D is attached to C above. "
= 7 components

・ Requirement enumeration method "~ A,
~ B arranged in the above A ,
Connected to said B, a C consisting of,
An apparatus characterized by comprising D attached to C above. "
Number of 4 components (8 components including components)

When components are not counted as in the requirement enumeration method, the number of underlined case components is the number of conditions necessary for realizing the proposition to be realized by each component. In the rewriting method, the number of underlined case components is a condition number necessary for a proposition to be realized by the title (device) of the invention.
Even in the requirement enumeration method, the object of the proposition of the number of case components to be obtained is not a component but an ultimate device. That is, in the requirement enumeration method, the device includes A, B, C, and D. Then, it should be considered that each of A, B, C, and D realizes each proposition. When it is defined that "a device has A, B, C, and D", the conditions necessary for A, B, C, and D to realize the proposition that the subject of the "device" is "provided". Therefore, it becomes a case component.
For this reason, even in the requirement enumeration method, even if each component is counted as a case component, it is logically consistent. In addition, differences due to differences between the requirement enumeration method and the rewriting method can be eliminated.

(3-1-2) Pattern / requirement enumeration method of method invention
A B step of adding P to O produced in the A step ,
A C step of heating the Q obtained in the B step with R ,
A step of extracting the heated S in the step C).
Yellow: 8 Green: 4 Total 12

・ Write-through method "Add P to O produced in process A ,
Heating the Q obtained in the step B with R ,
Extracting the heated S in the step C).
Yellow: 8 Green: 0 Total 8

  The constituent elements (A to D steps) are merely names, and do not particularly limit the matters specifying the invention. If components are included in the case components, the differences between the methods will be noticeable.

  From the above discussion, components (processes and steps) appearing in the method invention may not be included in the case component. Further, components (processes and steps) appearing in the method invention may be included in the case component. Further, the components appearing in the invention of the product may be counted by being included in the case component. In addition, the components appearing in the invention of the product may not be included in the case component.

  A product must always include components even when converting the component requirement enumeration system to the write-down system (the components of the product invention are visibly embodied). The method is sufficient for converting from the constituent requirement enumeration method to the writing-through method without inserting a process name or a step name (because each step of the method is an invisible conceptual concept. Does not appear in

  (3-2) The title of the invention (= name appearing at the end of the claim) is not counted as a case component.

(4) Rule of double definition If the invention-specific matter already defined in the previous stage or the processing operation defined in the previous stage is counted again in the subsequent stage, double counting will result. Therefore, it is not counted in the subsequent stage.

(4-1) action called "the registration registration by means" → said registration means already to register is a case where it ends in the preceding paragraph, without registration by the registration means, and registered by the registration means.

Another example) " Index entry including the key " has appeared in the first time → Include is a verb, and the case component is counted separately. Next, the " index entry including the key " appeared in the second time. Since the item that appeared last time has been received, it is not counted separately, but collectively as one rank component.

(4-2) "Imaging means for creating A and A created by the imaging means ~"
Regarding "A created by the imaging means" , when trying to count purely the number of case components, the part of "by the imaging means" has a morpheme "by". The “imaging means” can be said to be the number of case components that can be counted independently. However, the proposition of creating A by the imaging means has already been realized in the previous stage. For this reason, if "the imaging means" is counted again here, the proposition of creating A is realized again, and the proposition is realized twice. Since the number of case components as a parameter corresponding to the feasibility of the proposition deviates from the purpose, here, the phrase "created by the imaging means " is combined with "A" to form one case component number. I do.

(4-3) "and the number of printed sheets calculating means for calculating the number of printed sheets of printed material, - the number of printed sheets calculated by the number of printed sheets calculating means"
As for "from the number of printed sheets calculated by the above-described number-of-printed-sheets calculating means", the morpheme "by" is present in the part "by the above-mentioned number-of-printed-sheets calculating means" when purely trying to count the number of components. Therefore, it can be said that the “number of printed sheets calculating means” is a case component number that can be counted independently. However, the proposition that the number of prints is calculated by the print number calculator has already been realized in the previous stage. For this reason, if "the number of printed sheets calculating means" is counted again here, the proposition that the number of printed sheets is calculated is realized again, so that the proposition is double realized. Become. In this case, the word “calculated by the above-described print number calculation means” is combined with “the number of prints” to make one case component, because it deviates from the meaning of the number of case components as a parameter corresponding to the feasibility of the proposition. Number.

(5) Adverb rules (5-1) Adverbs directly related to verbs are not counted.
"In ~, the ~ freely, to be ~, so as not to ~, ~ not the" example like, Also, "move faster than ~" adverb "fast for receiving relates the verb" move "as Does not count as well. Further, there is also a case where the function as an adverb is expressed in a single word such as "integrally" without using a word such as "to".

(5-2)-As described above, for example, in the pattern of "(verb) like (verb)" (instructing to stop EX), it is not counted as an adverb.

(6) Rules for gerunds Gerunds are integrated with verbs and are not counted as case components. Examples of gerunds here are as follows.

(6-1) Agree with A = Agree to do A. For example, it is consent to control → control. In other words, because it is a gerund, no case component is counted at the place of "control."

(6-2) Performing ●● = agree to do.
Communicate → agree to communicate. The case component is not counted at "communication". Control → gerund (because it is synonymous with control). In addition to “do”, there is “execute” and the like.

(6-3) What can express an action in one word, for example, "cooperation", "action", and the like. By collaborating = By collaborating, we do not count because we agree. By opening / closing valve operation → By operating the opening / closing valve, we agree with it, so we do not count. " By collaboration " is synonymous with "by collaboration ." Cooperation is a noun and usually constitutes a case component. However, if this is defined as the latter, it is not counted as a case component.
In order to eliminate such an error, in the case of a gerund (a morpheme such as), counting is not performed. A gerund also means a noun that can express a movement such as collaboration in one word.

(7) Rules for adjectives Adjectives related to nouns and modifiers that modify nouns are considered integrated with nouns, and are not particularly counted independently.
For example, “ a large handle of” or “ information that can be input by using”. Since both are related to nouns, they are collectively referred to as one case component.
For example, the adjectives " removable storage medium " and "-possible" are related to the noun "storage medium", and such adjectives are combined with the noun to have one case component.
For example, “like (noun)” → this pattern also includes a noun and an adjective as one case component.
For example, "+ state verb (including adjectives)" In this case, it is divided once at the state verb. For example, “ Driving pulse signals with two different frequencies ”

(8) Each rule 1 of morpheme (morpheme to be divided)
The following A and B are nouns or noun phrases. Also, the meanings of “separate” and “separate” mean that A and B are not combined into one case component, but are extracted as A and B independent case components and counted as two case components. is there.

(8-1) “to”, “to”, “to”, “to” → Principal morpheme that always separates.
(8-2) "By A , ~" → Case component indicating method and means, divided and counted as one case component.
(8-3) "to", "to"
Since it becomes a case component that determines the target and the destination where the action is performed, it is divided. It is synonymous with "to"->"to", and is not an internal thing like "as" or "of", but has an effect on other components. For this reason, it is divided.
(8-4) “ B used for A ” → a case component indicating a use. In this case also, it is divided.
(8-5) " B indicating A " → Divide without grouping. Because, "show" is an agreement to display and is also a verb.
(8-6) “By” → case component representing means is divided and counted as one case component.
(8-7) “For” → case component representing the purpose is divided and counted as one case component.
(8-8) "B for A" is the same as above. “ Noun phrase for noun phrase ” → for = case component representing purpose Two case components. "Noun phrase for verb" → counts only noun phrases, not verbs. For this reason, it is a 1st rank component. For example, the number of case components is not counted at "controller for control" → "for control".
(8-9) “between” and “case component” that specifies something between the two. It is divided and counted as one case component. For example, “C between A and B” → A, B, and C each independently count the number of case components, in this case, the number of case components.
(8-10) " B including A " → A and B are separated. (Not included in extra components)
(8-11) Since it is the same rank as " B with A " → "to", it is divided. For example, correspondence between the elements of the previous SL element and the first display document of the structured document,
C → A, B, C each count of A and B (8-12) "of", "B from A", is "composed of A B" → the start and end points, or the raw material shown rated component Divide from.
(8-13) Because we agree with “at” and →, we divide.
(8-14) According to “ B accompanying A ” →, we agree with and separate.

(9) Each rule of morpheme (morpheme to be integrated)
The following A and B are nouns or noun phrases. In addition, the meaning of “integrate” means that A and B are not extracted as independent case components, but A and B are combined and extracted as one case component number.

(9-1) "To"
In most cases, they are stuck to "case" and "time" in (1). It is OK if integrated with these.
In other cases, it may be directly attached to a noun, such as "to the above-mentioned means," or "to the attachment body". In such a case, "ni" may be translated as "ni", and a noun to which "ni" is attached is extracted as a case component.
(9-2) “To”
Same as “to” → “of”. For example, it can be replaced with a terminal of the terminal → opponent to opponent. In other words, since it is apposition with "and to" → "of", without the terminal to the opponent this, and terminals that opponent.
(9-3) "B related to A" → A, B collectively as one rank component (agree with ∵ ~).
(9-4) “B related to A” → A, B collectively constitutes one rating component (agrees with ∵-).
(9-5) “B in A” → A and B are grouped into one case component (agreement with ∵) → Similarly, the case is expressed as “at”: (place, state).
(9-6) "B which becomes A" → A, B are collectively one rank component (agrees with ① ~). For example, an orthogonal transformation coefficient of an image to be searched → “becomes” is equivalent to “as”. Since “as” is considered as a single item and is not separated, the same applies here without dividing.
(9-7) "B which is A" → A, B are collectively one rank component (agree with ∵ ~). " ~ Part information which is information " → 1 rank component. It is simply a restatement that "it is part information", and there is no particular limitation on components.
(9-8) "B at A" → A, B collectively as one rank component (agrees with ∵ ~).
(9-9) The idea that “(of) B of A” → “of” and “in” is equal. A, B are collectively one rank component. However, when a verb form is included between A and B, the division is performed. For example, the position of an area to be viewed in the entire virtual screen of the electronic information .
(9-10) "B as A" → "No", A, B are collectively one rating component.
(9-11) “B related to A” → A, B are collectively one rating component.

(10) AND OR Rule (10-1) AND
A and B , A and B, etc. → Count case components of A and B respectively (10-2) OR
A or B, A or B, A or B, any one of A or B, A, B, etc. → A, B collectively the number of rank components.
-Any two of A, B, and C → A, B, and C collectively represent the number of two components. (∵Because at least two case components are selected, the number of case components is 2).

Implementation method of the present invention

  In carrying out the present invention, a personal computer (PC) is used, and the processing is performed based on a program read into the personal computer (PC).

  First, a character string described in a specific column (described in claims) in the electronic data-converted patent specification is extracted from the description of the electronic data-converted patent specification. Further, this specific column corresponds to one claim in the claims.

  Next, the number of case components is counted from the extracted character string. The number of case components means the number of case components in one claim. In the case of a main claim, the number of case components may be counted as it is from the character string extracted from the claim in which the main claim is defined, but in the case of a subclaim, a claim in which the subclaim is defined The number of case components is counted as it is from the character string extracted from, and the number of case components of the main claim to which the subclaim depends is added thereto.

  Then, the counted number of case components is displayed on a display or the like as the degree of limitation of the invention.

  At this time, the case component may be extracted using morphemes as shown in Table 1 as a mark. Incidentally, Table 1 here is merely an example, and it is needless to say that counting may be performed based on any other morpheme.

  The extraction of the case component may be performed based on the second embodiment.

  At this time, the above-mentioned count is to extract a word in which an overlapping operation is defined in the extracted character string, count the number of duplicate case components included in the extracted word, and calculate the number of duplicate case components from the counted total number of case components. A value obtained by subtracting the number of overlapping case components may be displayed as the degree of limitation of the invention. This takes into account the double counting described above.

  In the present invention, a character string described in a specific column in a patent specification before correction is extracted, and a character string described in a specific column in a patent specification after correction is extracted. The total number of components is counted from the character string extracted from the previous patent specification, and the character string extracted from the corrected patent specification, respectively. Alternatively, the difference from the total number of components in the corrected patent specification may be obtained, and this may be displayed as the above-described correction degree in the claims.

  Further, the present invention may be embodied as an apparatus such as a PC in which the program for performing the above-described analysis display is installed, or a program for causing a PC to execute the program, or a recording medium on which the program is recorded. It may be embodied as

  Further, the present invention may be applied to a network system. First, on the server side, data obtained by digitizing the number of case components in the patent specification in advance is stored in the server. Then, when a request for transmission of the number of case components for a desired patent is issued from the client side, the number of case components of the patent may be read from the server and transmitted to the client side via a network.

  If the number of case components has not yet been obtained for the patent requested to be transmitted from the client, the server counts the number of case components for the requested patent and sends this to the client via the network. You may send it.

  In extracting the case component, any existing text mining technology, data mining technology, linguistic analysis processing technology, or the like may be used. Then, using these techniques, it is determined whether or not each character string to be analyzed corresponds to a case component based on the above-described rules, and finally a case component for one claim is specified. Then, the total number of the specified case components for one claim is obtained, and this is output as the number of case components of the claim.

  In the present invention, not only the number of case components for each claim is output, but also the output data may be added to any information such as a patent map, a graph, and other evaluation values.

  In the present invention, the case component specified for each claim may be displayed on a screen, printed, or converted into data. That is, as shown in the following embodiments, underscores, highlights, and the like may be displayed so that the case component specified for the claim can be visually grasped.

  Also, in the present invention, conditions (case components) that must be cleared for realizing a proposition by a verb, as shown in FIG. 1, are typically displayed on a screen, printed, or converted into data. It may be. In FIG. 1, components are drawn, and arrows starting from the components are drawn (hereinafter, these arrows are referred to as basic arrows). Then, the condition for starting the operation of this component is indicated by an arrow that merges with the basic arrow. This condition corresponds to a case component, but an arrow drawn from this case component so as to merge with the basic arrow is called a tributary arrow. As a result, a skeleton-like figure is drawn in which the tributary arrow extending from the case component joins the basic arrow extending from the component. A verb may be written at the end of the basic arrow.

  Further, as another example, it is assumed that the component “information storage means” is defined by the following C-1).

  C-1) “Information storage means for storing discrimination information for discriminating an area in which image data in which an output time closest to the reference time is described is stored in the hard disk,

  C-1), which defines this information storage means, is seemingly complicatedly defined, but the cause is that the discrimination information, which is the case component of the verb "store" directly related to the information storage means, includes: In addition, the verb “discriminate” is affected, and the case component is affected by the verb “store”, and the case component image data is affected by the verb “describe”. This is because it contains four verbs in total.

  As described above, in the present invention, it is possible to display, display, and the like according to the same rule in any of the claims. At this time, the number of case components may be simultaneously described.

  Note that the present invention is not limited to the case component extraction described in the claims of the patent specification, and is also applicable to the case where case components are extracted based on the same method for any other documents. be able to.

Implementation method of the present invention

  In carrying out the present invention, a personal computer (PC) is used, and the processing is performed based on a program read into the personal computer (PC).

  First, a character string described in a specific column (described in claims) in the electronic data-converted patent specification is extracted from the description of the electronic data-converted patent specification. Further, this specific column corresponds to one claim in the claims.

  At this time, the case component may be extracted using morphemes as shown in Table 1 as a mark. Incidentally, Table 1 here is merely an example, and it is needless to say that counting may be performed based on any other morpheme.

  The extraction of the case component may be performed based on the second embodiment or the third embodiment.

  At this time, the above-mentioned count is to extract a word in which an overlapping operation is defined in the extracted character string, count the number of duplicate case components included in the extracted word, and calculate the number of duplicate case components from the counted total number of case components. A value obtained by subtracting the number of overlapping case components may be displayed as the degree of limitation of the invention. This takes into account the double counting described above.

  Further, the present invention may be embodied as an apparatus such as a PC in which the program for performing the above-described analysis is installed, or as a program for causing the PC to execute the program or a recording medium on which the program is recorded. It may be embodied.

  Further, the present invention may be applied to a network system. First, on the server side, data obtained by digitizing the number of case components in the patent specification in advance is stored in the server. Then, when a request for transmission of the number of case components for a desired patent is issued from the client side, the number of case components of the patent may be read from the server and transmitted to the client side via a network.

  In extracting the case component, any existing text mining technology, data mining technology, linguistic analysis processing technology, or the like may be used. Then, using these techniques, it is determined whether or not each character string to be analyzed corresponds to a case component based on the above-described rules, and finally a case component for one claim is specified.

  In the present invention, the case component specified for each claim may be displayed on a screen, printed, or converted into data. That is, as shown in the following embodiments, underscores, highlights, and the like may be displayed so that the case component specified for the claim can be visually grasped.

  Also, in the present invention, conditions (case components) that must be cleared for realizing a proposition by a verb, as shown in FIG. 1, are typically displayed on a screen, printed, or converted into data. It may be. In FIG. 1, components are drawn, and arrows starting from the components are drawn (hereinafter, these arrows are referred to as basic arrows). Then, the condition for starting the operation of this component is indicated by an arrow that merges with the basic arrow. This condition corresponds to a case component, but an arrow drawn from this case component so as to merge with the basic arrow is called a tributary arrow. As a result, a skeleton-like figure is drawn in which the tributary arrow extending from the case component joins the basic arrow extending from the component. A verb may be written at the end of the basic arrow.

  Next, a learned model is constructed by learning these extracted case components. A data set is sequentially created from the extracted case components and the verbs to which they are related. The method of extracting the case component is as described above, and the extraction of the verb it depends on may use text mining technology, for example, follow morphemes, or may be based on the case component extraction method described above. The extraction may be performed by any method or any conventional text mining method.

  By the way, the object to be extracted will be described by taking the description of the claims of the patent specification as an example, but is not limited thereto. It may be a target.

  If "a signal generating means for generating a drive signal in response to a request from the user, ...", the case components "in response to a request from the user" and "the drive signal" are extracted as the above case components. At the same time, it extracts the verb “generate” that it depends on.

  As a result, the case component (“in response to a request from the user”, “the driving signal”) and the verb (“generate”) are learned as a data set. This learning may use a well-known artificial intelligence technology. This data set is composed of one or more case components for one verb.

  Such data sets are extracted and learned through numerous patent specifications (including those disclosed through patent publications). As a result, a learned model composed of a neural network as shown in FIG. 3A is generated. In this example, the input side of the neural network is a verb, and the output side is a case component. However, the present invention is not limited to this, and the input side of the neural network may be a case component and the output side may be a verb.

  As input-side parameters, verbs A, B, C,... Are connected as output-side parameters to case components 1, 2, 3, 4,. In each node constituting the neural network, for example, as shown in FIG. 3B, the verb Aa and the case component 11 and the case component 12 are connected to each other via a degree of association. According to this example, for verb Aa, it can be considered that case component 11 is associated with association degree W1 and case component 12 is associated with association degree W2. If the association degree W1 is 70 and the association degree W2 is 30, the verb Aa is associated with the case component 11 with a higher relation and is associated with the case component 12 with a lower relation. Can be considered. Such a degree of association is naturally changed by learning the data set sequentially, since it is a node of the neural network.

  After the trained model is created in this way, a search is performed with reference to the trained model. In this search, first, a character string input is accepted. If the input character string is a verb, a case component is searched for based on the degree of association of the input verb. At this time, the case component may be searched for from the case component having a high degree of association, or may be proposed from the case component having a low degree of association.

  Conversely, in the course of learning, as shown in FIG. 3C, two or more verbs Aa and Ab may be linked to one node, and each may be weighted by the degree of association. is there. In such a case, for example, if the input character string is a case component, a search is made for a verb based on the input case component. At this time, a search may be made based on a verb having a high association degree W3 or W4, or a case component may be proposed based on a case component having a low association degree W3 or W4.

  By learning the data set of the verb and the case component in this way, it is possible to improve convenience in actually creating the claims automatically. For example, when it is desired to search for a case component frequently related to a verb, the case component can be proposed by searching with reference to the above-described learned model. As a result, it is possible to narrow down the candidates of the case component related to the verb in the claims, and it is possible to improve the convenience even if the user is prompted to confirm the last part. Similarly, when it is desired to search for a verb whose case component is frequently related, it can be proposed by searching with reference to the above-described learned model. As a result, it is possible to narrow down the candidates of the verb to which the case component depends in the claims, and to improve the convenience even if the user is prompted to confirm the last part.

  The construction of the learned model is not limited to the above-described embodiment. As shown in FIG. 4A, a phrase structure having a verb and one or more case components related to the verb and a proposition corresponding thereto may be sequentially learned as a data set.

  The method of extracting the case component is as described above, and the extraction of the verb it depends on may use text mining technology, for example, follow morphemes, or may be based on the case component extraction method described above. The extraction may be performed by any method or any conventional text mining method. In this way, a phrase structure having a verb and one or more case components related thereto is created. In this phrase structure, a proposition can be realized by combining verbs and case components as shown in FIGS. Therefore, if the phrase structure is determined, the proposition will be determined accordingly. A data set is created with such a phrase structure and proposition.

  The proposition extraction method is detected, for example, from the description in the detailed description in the patent specification. For example, assuming that there is a phrase structure composed of a case component "in response to a request from the user" and "a drive signal" and a verb "generate" that it depends on, It is assumed that the definition is made to realize one proposition of “use by drive signal”. This proposition is often described in patent specifications, for example, in the patent specification, "a drive signal can be used by generating a drive signal in response to a request from a user" and the like. Often described. In such a case, text information such as "drive signal" and words including "can be", "become possible", and "achieved" at the end of a sentence containing the text information can be fully utilized using text mining technology. Extracting. Then, from the extracted text information, a proposition “using a drive signal” is picked up.

  In this way, a text mining technique is used on the basis of the case component and the verb included in the phrase structure, a proposition corresponding to this is extracted, and a data set is created. Incidentally, the text data corresponding to the proposition is not limited to the automatic extraction through the text mining technique, and may be manually input by the user.

  Incidentally, the object to be extracted will be described with reference to the claims in the patent specification and the description of the embodiment (detailed description) as an example, but the present invention is not limited thereto, and all descriptions in the patent specification, and Any document other than the patent specification may be targeted.

  Such data sets are extracted and learned through numerous patent specifications (including those disclosed through patent publications). As a result, a learned model including a neural network as shown in FIG. 4A is generated. In this example, the input side of the neural network has a phrase structure (composed of a combination of a verb and a case component), and the output side is a proposition. However, the present invention is not limited to this, and the input side of the neural network may be a proposition and the output side may be a phrase structure.

  The input parameters include a phrase structure composed of a combination of verb A and case components 1 and 2, a phrase structure composed of a combination of verb B and case component 3, and a combination of verb C and case components 4 and 5. The clause structure... Is connected to the propositions α, β, γ... Via the neural network as output-side parameters. In each of the nodes constituting the neural network, for example, as shown in FIG. 4B, a phrase structure Ba and propositions α21 and α22 are connected to each other via a degree of association. According to this example, for the phrase structure Ba, it can be considered that the proposition α21 is associated with the association degree W1 and the α22 is associated with the association degree W2. If the association degree W1 is 70 and the association degree W2 is 30, the phrase structure Ba is associated with the proposition α21 with a higher relation, and is associated with the proposition α22 with a lower relation. Can be considered. Such a degree of association is naturally changed by learning the data set sequentially, since it is a node of the neural network.

  After the trained model is created in this way, a search is performed with reference to the trained model. In this search, first, a character string input is accepted. Extract case components and verbs from the input character string. This extraction method is as described above. Alternatively, if a character string is input independently with a case component and a verb associated with each other, it may be used as it is. A proposition based on the degree of association corresponding to the phrase structure composed of the case component and the verb extracted from the character string input in this way is searched for. At this time, a search may be made from a proposition with a high degree of association, and a proposal may be made from a proposition with a low degree of association.

  Conversely, in the course of learning, as shown in FIG. 4C, there are cases where two or more phrase structures Ba and Bb are linked to one node, and each is weighted by the degree of association. There is. In such a case, for example, when the input character string is a proposition, a phrase structure based on the input proposition is searched. At this time, a search may be made from a phrase structure having a high degree of association W3 or W4, or a suggestion may be made by searching from a phrase structure having a low degree of association W3 or W4.

  By learning the phrase structure and the data set of the proposition in this way, it is possible to improve convenience in automatically creating claims automatically. For example, when it is desired to search for a proposition that frequently depends on the phrase structure, the search can be proposed by referring to the above-described learned model. As a result, proposition candidates corresponding to the phrase structure can be narrowed down in the claims, and the convenience can be improved even if the user is prompted to confirm the last part. Similarly, when it is desired to search for a phrase structure from a proposition, it can be proposed by searching with reference to the above-described learned model. As a result, it is possible to narrow down the phrase structure candidates necessary for realizing the intended proposition in the claims, and to improve the convenience even if the user is prompted for confirmation at the end. it can.

  The construction of the learned model is not limited to the above-described embodiment. As shown in FIG. 5A, a dependency structure having a verb and one case component related to the verb, and a morpheme to be dependent on the verb may be sequentially learned as a data set.

  The method of extracting the case component is as described above, and the extraction of the verb it depends on may use text mining technology, for example, follow morphemes, or may be based on the case component extraction method described above. The extraction may be performed by any method or any conventional text mining method. In this way, a dependency structure having a verb and one or more case components related thereto is created. The morphemes to be dependent in this dependency structure may be, for example, the morphemes shown in Table 1 as a mark, or may be obtained by making full use of a known text mining technique. A data set is created using such a dependency structure and a morpheme.

  Incidentally, the object to be extracted will be described with reference to the claims in the patent specification and the description of the embodiment (detailed description) as an example, but the present invention is not limited thereto, and all descriptions in the patent specification, and Any document other than the patent specification may be targeted.

  Such data sets are extracted and learned through numerous patent specifications (including those disclosed through patent publications). As a result, a learned model including a neural network as shown in FIG. 5A is generated. In this example, the input side of the neural network has a dependency structure (composed of a combination of a verb and a case component), and the output side has a morpheme. However, the present invention is not limited to this, and the input side of the neural network may have a morpheme and the output side may have a dependency structure.

  The input parameters include a dependency structure composed of a combination of verb A and case component 1, a modification structure composed of a combination of verb B and case component 2, and a modification structure composed of a combination of verb C and case component 3. Structures... Are connected on the output side to morphemes a, b, c,. As shown in FIG. 5B, for example, the individual nodes constituting the neural network have a dependency structure Ca and morphemes a31 and a32 connected to each other via a degree of association. According to this example, with respect to the dependency structure Ca, it can be considered that the morpheme a31 is associated with the degree of association W1 and the morpheme a32 is associated with the degree of association W2. If the degree of association W1 is 70 and the degree of association W2 is 30, the dependency structure Ca is associated with a higher relationship with the morpheme a31 and is associated with a lower relationship with the morpheme a32. Can be considered. Such a degree of association is naturally changed by learning the data set sequentially, since it is a node of the neural network.

  After the trained model is created in this way, a search is performed with reference to the trained model. In this search, first, a character string input is accepted. Extract case components and verbs from the input character string. This extraction method is as described above. Alternatively, if a character string is input independently with a case component and a verb associated with each other, it may be used as it is. The morpheme based on the degree of association corresponding to the dependency structure composed of the case component and the verb extracted from the character string input in this way is searched for. At this time, a search may be made from a morpheme having a high degree of association, and a suggestion may be made from a morpheme having a low degree of association.

  Conversely, in the course of learning, as shown in FIG. 5C, two or more dependency structures Ca and Cb may be linked to one node, each of which is weighted by the degree of association. There are cases. In such a case, for example, when the input character string is a morpheme, a dependency structure based on the input morpheme is searched. At this time, a search may be made from the dependency structure having a high association degree W3 or W4, or a proposal may be made by searching from a dependency structure having a low association degree W3 or W4.

  By learning the dependency structure and the data set of the morpheme in this way, it is possible to improve convenience in automatically creating the claims. For example, when it is desired to search for a morpheme that frequently depends on the dependency structure, the search can be proposed by referring to the learned model described above. As a result, in the claims, morpheme candidates corresponding to the dependency structure can be narrowed down, and even if the user is prompted to confirm the last part, the convenience can be improved. Similarly, when it is desired to search for a dependency structure from a morpheme, it can be proposed by searching with reference to the learned model described above. As a result, it is possible to narrow down candidates for a dependency structure necessary for realizing an intended morpheme in the claims, and to improve the convenience even when prompting the user to confirm at the end. Can be.

Claims (6)

Learning to create a trained model in which the relationship between a verb and a case component is defined as three or more levels of association by sequentially learning the verb extracted from the character string and the case components involved in the verb as a data set. A case component learning program characterized by causing a computer to execute steps. A character string input receiving step for receiving a character string input;
With reference to the learned model created in the learning step, if the character string input in the character string input receiving step is a verb, a case component based on the degree of association of the input verb is searched for. And a search step for searching for a verb based on the degree of association of the input case component when the character string input in the character string input receiving step is a case component. The described case component learning program. By sequentially learning a phrase structure having a verb and one or more case components related thereto and a proposition corresponding thereto as a data set, the relation between the phrase structure and the proposition can be three or more degrees of association. A case component learning program characterized by causing a computer to execute a learning step of creating a trained model defined as: A character string input receiving step for receiving a character string input;
With reference to the trained model created in the learning step, if the character string input in the character string input receiving step has a phrase structure, search for a proposition based on the degree of association of the input phrase structure. 4. The method according to claim 3, further comprising the step of: when the character string input in the character string input receiving step is a proposition, searching for a phrase structure based on the degree of association of the input proposition. The described case component learning program. By sequentially learning a dependency structure of a verb and a case component related to the verb and a morpheme to be changed as a data set, the relationship between the dependency structure and the morpheme has three or more degrees of association. A case component learning program characterized by causing a computer to execute a learning step of creating a trained model defined as: A character string input receiving step for receiving a character string input;
Referring to the trained model created in the learning step, if the character string input in the character string input receiving step has a dependency structure, a proposition based on the degree of association of the input dependency structure And if the character string input in the character string input receiving step is a morpheme, a search step of searching for a dependency structure based on the degree of association of the input morpheme is provided. The case component learning program according to claim 5.

Images