KR101890207B1 - Method and apparatus for named entity linking and computer program thereof - Google Patents

Abstract

The present invention relates to an object name connection method, an apparatus and a computer program, and more particularly to a method for connecting an object name to a multiresistant word whose object name connection system is located in a text including one or more sentences, A polyphonic output step of calculating a word; A neural network selecting step of selecting a neural network corresponding to the multiple words among a plurality of neural networks; A neural network input step of inputting all or part of words included in the text into the selected neural network; And an object name connecting step of connecting an object name to the multilingual word in consideration of an output of the neural network, wherein the plurality of the neural networks are generated individually for a plurality of plural words. do.

Description

METHOD AND APPARATUS FOR NAMED ENTITY LINKING AND COMPUTER PROGRAM THEREOF FIELD OF THE INVENTION [0001]

The present invention relates to an object name connection method, an apparatus, and a computer program. More particularly, the present invention relates to a method for performing an object name connection by grasping an exact meaning of a used word using a plurality of neural networks, Device and a computer program.

Named Entity Linking (Named Entity Linking) is a method for determining the meaning of a word or the like used in a given document when the word or the like is interpreted in a given document, (For example, judging whether the word "spider" used in the sentence means a singer or an arthropod) is used to connect the word or the like to a person, object, place or the like. The entity name association can be an important component in a search engine, a conversation system, etc. in which the result of semantic judgment of a dominant word affects quality.

Conventionally, a feature is calculated in the vicinity of a connection target word of a given document, and then a connection of an object name is performed by applying a support vector machine (SVM), a logistic regression analysis, a similarity analysis, Respectively. However, in the case of the related art, when the number of distinct object names is increased, it may become difficult to obtain an exact result with respect to time with an increase in the required time. Particularly, in a search engine in which the object name connection is not limited to a small number of specific words but a plurality of terms are used, the distinction object name can be used from 100 to 10,000 units, The problem can be followed.

On the other hand, it is possible to consider a method of processing a plurality of object names by using a deep learning technique, which has recently been attracting attention in fields such as natural language processing, image analysis, and speech recognition. In such a case, As the number of users increases, the amount of computation can increase exponentially. As a result, the required amount of computing resources can be greatly increased, and the cost for operating the entity name connection also increases. Furthermore, when using the deep learning technique as described above, even if a specific object name is connected incorrectly due to a complicated model structure, it is difficult for an administrator to identify the cause of the object and correct the result.

Accordingly, even if the number of distinguished object names is increased, it is possible to perform an accurate object name connection, and also it is possible to effectively suppress the increase of the computation resources by processing the object name connection efficiently. Further, Although it is required to provide an object name connection technique capable of correcting this problem, there is not yet a proper solution for this problem.

Korean Patent No. 10-1255957 (Apr. 24, 2014)

It is an object of the present invention to provide an object name connection method, an apparatus, and a computer program capable of performing an accurate object name connection even when the number of distinguished object names increases, in order to solve the above- do.

It is another object of the present invention to provide an object name connection method, an apparatus, and a computer program capable of efficiently processing an object name connection even when the number of object names is increased, thereby suppressing an increase in the amount of computation resources required.

Furthermore, it is an object of the present invention to provide an object name connection method, an apparatus, and a computer program capable of appropriately correcting even a wrong object name.

According to an aspect of the present invention, there is provided a method for connecting an object name to a multinomial word in which an object name connection system is located in a text including at least one sentence, A polynomial calculation step of calculating a polynomial; A neural network selecting step of selecting a neural network corresponding to the multiple words among a plurality of neural networks; A neural network input step of inputting all or part of words included in the text into the selected neural network; And an object name connecting step of connecting an object name to the polyphonic word in consideration of an output of the neural network, wherein the plurality of neural networks are individually generated for a plurality of polyphonic words.

According to another aspect of the present invention, there is provided a method of processing a text transmitted from a terminal, the method comprising: a text transmission step in which a server receives text including one or more sentences from a terminal; Calculating a plural word in the text; A neural network selecting step of selecting a neural network corresponding to the multiple words among a plurality of neural networks; A neural network input step of inputting all or part of words included in the text into the selected neural network; A semantic determination step of determining the meaning of the ambiguous word in consideration of the output of the neural network; And an output transmission step of transmitting, to the terminal, an output calculated for the text in consideration of the meaning of the ambiguous word, wherein the plurality of neural networks are generated individually for a plurality of the plural words.

A computer program according to another aspect of the present invention is characterized by being a computer program stored in a computer-readable medium for causing a computer to execute each of the steps of the object name connection method or the text processing method described above.

According to another aspect of the present invention, there is provided an object name linking system for an object name linking system for a multilingual word located in text including at least one sentence, the system comprising: a genetic term calculating unit for calculating a multilingual word within the text; A neural network selecting unit for selecting a neural network corresponding to the multiplex word among a plurality of neural networks; A neural network input unit for inputting all or part of words included in the text by the selected neural network; And an object name connection unit for connecting the object name to the multilingual word in consideration of the output of the neural network, wherein the plurality of the neural networks are individually generated for each of the plurality of multilingual words.

According to another aspect of the present invention, there is provided a server for processing text transmitted from a terminal, the server comprising: a text transmission unit for receiving text including one or more sentences from the terminal; A polygonal word calculation unit for calculating a polygonal word located in the text; A neural network selecting unit for selecting a neural network corresponding to the multiplex word among a plurality of neural networks; A neural network input unit for inputting all or part of words included in the text by the selected neural network; A meaning judging unit for judging the meaning of the ambiguous word in consideration of the output of the neural network; And an output unit for transmitting an output calculated for the text to the terminal in consideration of the meaning of the ambiguous word, wherein the plurality of neural networks are generated separately for a plurality of the plural words.

According to an embodiment of the present invention, by precisely understanding the meaning of a plural word by using a plurality of neural networks individually configured in units of plural words and performing object name connection, accurate object name connection is performed even if the number of the separated target object is increased Furthermore, it is possible to effectively suppress the increase of the computation resources required by processing the object name connection.

In addition, according to an embodiment of the present invention, the object name connection is performed using a plurality of shallow neural networks individually generated for each of the plural words, so that even if a specific object name is misconnected, it can be corrected appropriately.

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.
1 and 2 are views for explaining the operation of the object name connection system according to an embodiment of the present invention.
3 is a diagram illustrating an example of a user terminal for inputting a search term in an entity name concatenation method according to an embodiment of the present invention.
FIG. 4 is a flowchart of a method of connecting an entity name according to an embodiment of the present invention.
5 is an exemplary diagram of a web document to which entity names are linked according to an embodiment of the present invention.
FIG. 6 is a diagram illustrating an exemplary operation of the object name connection method according to an exemplary embodiment of the present invention. Referring to FIG.
7 is an exemplary diagram illustrating the structure of a neural network according to an embodiment of the present invention.
FIG. 8 is a diagram for explaining a process of selecting qualities for a plural word according to an embodiment of the present invention.
FIG. 9 is a diagram illustrating an example of applying a negative sampling technique in constructing a neural network according to an embodiment of the present invention. Referring to FIG.
FIG. 10 is a diagram illustrating an example of applying a weight according to a position of a feature in constructing a neural network according to an embodiment of the present invention. Referring to FIG.
FIG. 11 is an exemplary diagram showing the calculated accuracy of an object name connection method according to an embodiment of the present invention.
12 is a configuration diagram of a text processing system according to an embodiment of the present invention.
13 is an exemplary diagram illustrating a user terminal for inputting a search statement in a text processing method according to an embodiment of the present invention.
14 is a flowchart of a text processing method according to an embodiment of the present invention.
FIG. 15 is a configuration diagram of an entity name connection system according to an embodiment of the present invention.
16 is a configuration diagram of a text processing server according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention is capable of various modifications and various embodiments, and specific embodiments will be described in detail below with reference to the accompanying drawings.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

The terms first, second, etc. may be used to describe various components, but the components are not limited by the terms, and the terms are used only for the purpose of distinguishing one component from another Is used.

Hereinafter, exemplary embodiments of a method, a user terminal, a server, and a computer program for providing a translation according to the present invention will be described in detail with reference to the accompanying drawings.

First, FIG. 1 illustrates an operation of an entity name connection system 100 according to an embodiment of the present invention. 1, the object name connection system 100 according to an exemplary embodiment of the present invention is connected to a database 30 and stores a document (for example, a blog, A variety of electronic documents including cafes, news, and the like), object names are concatenated for the multilingual words located in the text including one or more sentences included in the document.

More specifically, for example, if a specific electronic document stored in the database 30 contains a sentence "You picked 'One Piece' today at your daughter and a store as shown in FIG. 2, The word 'One Piece' can be 'One Piece', which is the title of a cartoon, depending on its use, or 'One Piece', which is a type of clothes. On the other hand, in the object name concatenation system 100 according to the embodiment of the present invention, it is determined whether the 'one piece' means comic or clothes in the sentence, The object name is connected. Accordingly, object names such as 'One Piece _ Clothes' may be connected to 'One Piece' of the sentence and stored.

In the present invention, one word such as "One Piece" has a multiple meaning. However, the present invention is not necessarily limited thereto, and the present invention can be applied even when two or more words have a plural meaning (For example, "winter sea (song title / sea of a certain season)"). Accordingly, the phrase "word" in the present invention includes a phrase including two or more words.

In addition, the entity name connection system 100 according to an embodiment of the present invention can provide services to users by reflecting the entity name connection result. More specifically, for example, as shown in FIG. 3, when the user of the terminal 10a desires to perform a search for a cartoon 'One Piece' (for example, (In the case of inputting in a form such as a piece (cartoon) or a piece of clothes), the entity name connection system 100 according to an embodiment of the present invention may have a multi- It is possible to provide the search result considering the meaning of the search word so that the user can more conveniently obtain the desired search result.

2, the entity name concatenation system 100 may include one or more servers. The entity name concatenation system 100 may include one or more servers. A device configured by using dedicated hardware for performing entity name connection, and the like.

The terminals 10a and 10b may be various mobile terminals such as a smart phone, a tablet PC, a PDA, and a mobile phone. In addition, various terminals such as a personal computer (PC) and a notebook PC may be adopted.

The communication network 30 that connects the user terminals 10a and 10b and the object name connection system 100 may include a wired network and a wireless network. Specifically, the communication network 30 may include a local area network (LAN) , A metropolitan area network (MAN), a wide area network (WAN), and the like. In addition, the communication network 30 may include a known World Wide Web (WWW). However, the communication communication network 30 according to the present invention is not limited to the above listed networks, but may include at least a known wireless data network or a known telephone network or a known wire / wireless television network.

FIG. 4 illustrates a flowchart of an entity name connection method according to an embodiment of the present invention. As shown in FIG. 4, the object name connection method according to an embodiment of the present invention includes a method of connecting an object name to a dominant word located in a text including one or more sentences (S110) for calculating a complex word in the text, a neural network selection step (S120) for selecting a neural network corresponding to the complex word among the plurality of neural networks, a step of selecting a word (S130) for inputting all or a part of the plurality of neural networks, and an object name connecting step (S140) for connecting object names to the multiple words in consideration of the output of the neural network, It is generated separately for plural plural words.

Referring to FIG. 4, an object name connection method according to an embodiment of the present invention is divided into individual steps and examined in detail.

First, in step S110, a complex word is calculated in a text including one or more sentences. At this time, in step S110, it is possible to compute the complex word included in the text by comparing the text with a predetermined list of words.

For example, in the object name connection system 100 according to an embodiment of the present invention, a plurality of plural words to be used for connection of object names can be selected in advance and a list of plural words including the multiple words can be stored and managed. Accordingly, in the entity name concatenation system 100 according to an embodiment of the present invention, a compound name for performing an object name connection in the text may be calculated by comparing the compound name list with the text.

For example, the above-mentioned list may include a generic term such as "spider "," one piece ", etc. Accordingly, when text is given (for example, Quot ;, or " One Piece Season 8 Broadcasting has been started from yesterday "in FIG. 5 (b)), the polygraph can be calculated (" one piece ") from the text in comparison with the above-

Furthermore, the present invention does not necessarily use the above-mentioned plural language list, but can be applied without any particular limitations as long as it is possible to appropriately calculate a plural language included in a given text.

Next, in step S120, a neural network corresponding to the dominant word is selected from a plurality of neural networks. Particularly, in the entity name connection system 100 according to an embodiment of the present invention, there are individual neural networks corresponding to the respective plural terms of plural terms, and these plural neural networks constitute the plural neural networks. Accordingly, the plurality of neural networks may be neural networks individually generated for a plurality of verbal words.

6, in the object name concatenation system 100 according to an embodiment of the present invention, the plurality of neural networks may be classified into multiple words (e.g., "one piece" Spider "(singer / arthropod)," day "(unit of date / restaurant name), etc.), and when a plural word is generated to perform the object name connection, Can be selected.

More specifically, for example, if the plural word "one piece" is calculated from the text containing one or more sentences, the neural network (FIG. 6A) generated for the plural word " , And then the entity name association for the plural word "Piece" can be performed using the selected neural network.

In addition, in the entity name concatenation system 100 according to an embodiment of the present invention, a neural network for the ambiguous word can be selected using a predefined plural word list. That is, as described above, the entity name linking system 100 according to an embodiment of the present invention may include a list of plural terms, and the list of the plural terms may include information on the corresponding neural networks.

Accordingly, in the entity name concatenation system 100 according to an embodiment of the present invention, a neural network for the ambiguous word can be selected by using a predefined plural word list. However, the present invention is not limited thereto, and the present invention can be applied without any particular limitations as long as it is a method capable of appropriately selecting a neural network corresponding to the calculated polygonal language.

In addition, in the entity name concatenation system 100 according to an embodiment of the present invention, the plurality of neural networks may be a shallow neural network (concept corresponding to a deep neural network of deep learning) in which a hidden layer does not exist.

Accordingly, when the object name connection is performed using the support vector machine (SVM), the logistic regression, the similarity analysis technique, etc. according to the related art, However, in the object name concatenation system 100 according to an embodiment of the present invention, object name concatenation is performed using a neural network generated for each term, It is possible to effectively suppress the increase of the required time and the decrease of the accuracy even if the number of persons increases.

In addition, when performing the entity name connection using the Deep Learning method, the computational resources required for the computation increases dramatically as the computation amount increases exponentially with the increase of the names of the objects to be discriminated. The entity name concatenation system 100 according to an embodiment of the present invention may also perform entity name concatenation using a corresponding neural network for a plurality of terms located in text including one or more sentences , It is possible to effectively prevent the increase in the amount of computation resources required even if the number of object names to be discriminated increases.

Next, in step S130, all or part of the words included in the text are input to the selected neural network. For example, in Figure 5 (a), the words ("Today," " Daughter ", " Store "or the like), or input words within a predetermined distance (for example, about 15 words before and after) on the basis of the plural word" one piece ", or input all words included in the text Do.

Further, in the entity name concatenation method according to an embodiment of the present invention, a feature extraction step of extracting one or more features corresponding to the complex word in the text may be further included. At this time, in the neural network input step (S130), the extracted one or more qualities are input to the selected neural network.

More specifically, in extracting the qualities of the plural word "one piece" in the text of FIG. 5 (a), nouns, adjectives, verbs, etc. included in the text are selected and extracted as the qualities of the plural word " (For example, "Today", "Daughter", "Store", "Go", "Pretty", "Denim Dress" in FIG. 5 (a)).

Further, the feature may be extracted from words within a predetermined distance from the dominant word in the text. For example, in FIG. 5 (a), qualities can be extracted from words within a predetermined number of words before and after the "piece".

In addition, the feature may be extracted from words included in a sentence representing the text such as the title, summary, and source of the text.

Furthermore, the text may be extracted in comparison with a predetermined list of qualities for each of the plural words. For example, in the case where "DENIM Piece", "Wear", "Pretty", "Rupee", "Shanks" and the like are included in the predetermined qualities for the above-mentioned "Piece" "I went to the store with my daughter today and picked a dress," "It was more fun to see my daughter wearing a pretty denim dress," and "Denim dress," "Pretty," and "Dressed" It can be extracted with qualities.

In the entity name concatenation system 100 according to an embodiment of the present invention, the amount of computation required in the neural network can be reduced and the quality of the result can be effectively secured through the above-described specific qualities extraction techniques.

Finally, in step S140, the entity name is connected to the dominant word in consideration of the output of the neural network. More specifically, in the entity name concatenation system 100 according to an exemplary embodiment of the present invention, the meaning of the ambiguous word included in the text is determined by considering the type, number, and weight of the qualities included in the text , And thus can perform object name association for the plural word.

For example, in FIG. 7, the structure of a neural network according to an embodiment of the present invention is illustrated. 7, the neural network according to an embodiment of the present invention includes predefined qualities (denoted as "denim dress", "dressed" as qualities for the generic term "dress" in FIG. 7) , "Shuffling", "shuffling", "coordination", "luffy", and "shanks" are used as an example) The output value is calculated in consideration of the weight, and the meaning of the plural word is grasped.

For example, in the sentence of FIG. 5 (a) ("I went to the store with my daughter today and picked the dress, I found the daughter dressed in a pretty denim dress and I enjoyed it more"), Quot; denim dress "and" mouth "are taken into account in the neural network, and the output value is calculated at each output node. Accordingly, the output value at the "OUTPUT_Piece_Clothes" node of FIG. 7 becomes 0.12 + 0.20 = 0.32, and the output value at the "OUTPUT_Piece_Cartoon" node becomes 0.01 + 0.01 = 0.02 Accordingly, in the object name concatenation system 100 according to the embodiment of the present invention, it is possible to judge that the plural word "one piece" of the text is the meaning of clothes rather than a cartoon in consideration of the output value.

On the other hand, the sentence in Figure 5 (b) ("You've been waiting for a long time, I have started the one-piece season 8 from yesterday, and I am looking forward to the adventure of Rupee to lead the pirates, especially Nami and Shanks.") Quot; OUTPUT_Piece_Clothes "node has a value of 0.01 + 0.01 = 0.02 when calculating the output value at the node of the output terminal, And the output value at the "OUTPUT_Piece_Cartoon" node has a value of 0.24 + 0.18 = 0.32. Accordingly, in the object name connection system 100 according to the embodiment of the present invention, It is possible to judge that the "one piece" is not the clothes but the meaning of the cartoon.

In addition, as can be seen from FIG. 7, in the entity name concatenation system 100 according to an embodiment of the present invention, when a text including a plural word is given, In particular, if the connection of the object name to the plural words included in the specific text is performed improperly, the system administrator can clearly identify the cause of the error, It is possible to correct the error by adjusting the weight or adding or deleting the qualities.

In addition, FIG. 8 illustrates a process of selecting qualities for each DICOM according to an embodiment of the present invention. As can be seen from Fig. 8 (a), in selecting the qualities of a specific plural word (the plural word "Kim So Yeon (1980)" in Fig. 8A), words used to describe the plural word can be selected and used have. For example, in one embodiment of the present invention, in selecting qualities for a specific plural word, the words included in the document of the Wikipedia Wikipedia Encyclopedia (ko.wikipedia.org) are selected. Accordingly, as can be seen from FIG. 8 (b), the selected qualities for the above-mentioned "Korean Sojyon (1980)" include "35 years old" And the like.

However, the present invention is not necessarily limited to the above. In addition, the present invention is not limited to the above-mentioned embodiments. For example, if the method is a method capable of appropriately selecting qualities of a specific word such as a word reflecting a characteristic of a given word, Can be selected.

Further, in selecting the qualities of a specific plural word in the entity name concatenation system 100 according to an embodiment of the present invention, mutual information, chi-square, information gain technique And so on.

9, in the entity name concatenation system 100 according to an embodiment of the present invention, the qualities of a specific plural word can be obtained by using an information gain with negative sampling technique It can also be selected. As can be seen in Fig. 9 (a), certain similar words can be shared by the same qualities (Fig. 9 (a) with Kim So Yeon (1980) and Kim So Yeon (2002) Qualities), which can lead to overfitting problems. 9 (b), in the object name concatenation system 100 according to an embodiment of the present invention, the noise samples (Lee Ji Hoon _ (Sports), One Piece ), Etc.) are added to the neural network to solve problems such as overfitting and selection bias due to insufficient data.

Furthermore, in the object name concatenation system 100 according to an embodiment of the present invention, a learning context may be added through searching for news, blog, etc., with the selected qualities as seed features. Accordingly, it is possible to perform an operation of extending the qualities using the added learning context or the like.

In addition, in the entity name connection system 100 according to an embodiment of the present invention, in constructing and learning a neural network, weights are calculated and optimized for each selected feature. At this time, the weight can be calculated and optimized by applying a back propagation technique.

Further, in calculating the weight for each of the qualities, the weight may be calculated in consideration of a distance from a given vernacular word ("Kim So Yeon" in Fig. 10) to qualities as shown in Fig. For example, in Fig. 10, "couple" is a word located within a predetermined distance from the above-mentioned "Kim So Yeon" (local context in Fig. 10), "MBC" 10), a weight corresponding to the distance from the plural is given, and a weight for the qualities of the plural can be calculated and an optimum value can be derived.

FIG. 11 shows the calculated accuracy of the entity name concatenation system 100 according to an embodiment of the present invention. As can be seen from FIG. 11, in performing entity name association for a plural word included in the text, according to the prior art, (1) the case of connecting to the most used meaning (Baseline in FIG. 11) (63%), accuracy in the case of the random forest (Random Forest in FIG. 11) and the support vector machine (support vector machine in FIG. 11) were only 84.2% and 81.3%, respectively. 100), the accuracy was improved to 90.5%.

FIG. 12 is a diagram for explaining the operation of the text processing server 200 according to an embodiment of the present invention. 12, the text processing server 200 according to an exemplary embodiment of the present invention is connected to the terminals 10a and 10b of the user, Performs a search or the like in consideration of the meaning of the contained plural terms, and transmits the result to the terminals 10a and 10b.

More specifically, as can be seen from FIG. 13, when the user's input words include a plural word (in the case of "spider is an insect" in FIG. 13) The word 'spider' may be the name of the singer 'spider' according to its use, or it may be a kind of arthropod, 'spider'. On the other hand, in the text processing server 200 according to an embodiment of the present invention, the 'spider' in the sentence is judged whether it means a singer or an arthropod, and the searched word is searched. Accordingly, in the text processing server 200 according to an embodiment of the present invention, a search result considering a meaning of the plural word is provided for a search word including a plural word, so that a user can more easily obtain a desired search result .

Furthermore, the text processing server 200 according to an exemplary embodiment of the present invention is not limited to performing the search service. In addition, the text processing server 200 may perform various services for receiving text, processing texts, and calculating results.

The terminals 10a and 10b may be various mobile terminals such as a smart phone, a tablet PC, a PDA, and a mobile phone. In addition, various terminals such as a personal computer (PC) and a notebook PC may be adopted.

The communication network 30 that connects the user terminals 10a and 10b and the text processing server 200 may include a wired network and a wireless network. More specifically, the communication network 30 may include a local area network (LAN) A metropolitan area network (MAN), a wide area network (WAN), and the like. In addition, the communication network 30 may include a known World Wide Web (WWW). However, the communication communication network 30 according to the present invention is not limited to the above listed networks, but may include at least a known wireless data network or a known telephone network or a known wire / wireless television network.

FIG. 14 shows a flowchart of a text processing method according to an embodiment of the present invention. 14, the text processing method according to the embodiment of the present invention is characterized in that the text processing server 200 receives a text including one or more sentences from the terminals 10a and 10b, (S210), a polyphonic language calculation step S220 of calculating a polyphonic word in the text, a neural network selection step S230 of selecting a neural network corresponding to the polyphonic word among the plurality of neural networks, A meaning determination step S250 for determining the meaning of the polyphonic word in consideration of the output of the neural network, and a semantic determination step S250 for considering the meaning of the ambiguous word to the terminals 10a and 10b And an output transmission step (S260) of transmitting an output calculated for the text, wherein the plurality of neural networks are individually .

The text processing method according to an embodiment of the present invention may be implemented similarly to the case of the entity name connection method according to the embodiment of the present invention described above, There is a difference in that the output is calculated for the text by inputting all or a part of the text into the neural network corresponding to the calculated polygonal word to determine the meaning of the polygonal word, The detailed description of each step of the method can be easily implemented by referring to the entity name connection method according to the embodiment of the present invention described above, and a detailed description thereof will be omitted.

According to still another aspect of the present invention, a computer program is a computer program stored in a computer-readable medium for causing a computer to execute each of the steps of the object name connecting method or the text processing method. The computer program may be a computer program containing machine code generated by a compiler, as well as a computer program containing a high-level language code that may be executed in a computer using an interpreter or the like. At this time, the computer is not limited to a personal computer (PC), a notebook computer, or the like, and may include a central processing unit (CPU) such as a server, a smart phone, a tablet PC, a PDA, Device. The computer-readable medium may also be an electronic storage medium (e.g., ROM, flash memory, etc.), a magnetic storage medium (e.g., floppy disk, hard disk, etc.) A CD-ROM, a DVD, etc.), and the like.

In addition, FIG. 15 illustrates the configuration of the object name connection system 100 according to an embodiment of the present invention.

As shown in FIG. 15, the entity name connection system 100 according to an embodiment of the present invention includes a polyphony calculation unit 110, a neural network selection unit 120, a neural network input unit 130, As shown in FIG.

The object name connection system 100 according to an embodiment of the present invention is divided into individual components. The object name concatenation system 100 according to an embodiment of the present invention can be inferred from the description of the object name concatenation method according to the embodiment of the present invention described above, Is omitted.

First, the polygonal language calculation unit 110 calculates a polygonal word in a text including one or more sentences.

Then, the neural network selecting unit 120 selects a neural network corresponding to the above-mentioned plural words among the plurality of neural networks.

Then, the neural network input unit 130 inputs all or part of words included in the text by the selected neural network.

Finally, the object name connecting unit 140 connects the object name to the multinomial word in consideration of the output of the neural network.

Here, the plurality of neural networks is a neural network individually generated for each of a plurality of verbal words.

In addition, in the entity name concatenation system 100 according to an embodiment of the present invention, the plurality of neural networks may be a shallow neural network having a hidden layer within one layer.

In addition, FIG. 16 illustrates a configuration diagram of a text processing server 200 according to an embodiment of the present invention.

16, the text processing server 200 according to an embodiment of the present invention includes a text receiving unit 110, a plural word computing unit 120, a neural network selecting unit 130, a neural network inputting unit 140, A determination unit 150, and an output unit 160.

Hereinafter, the text processing server 200 according to an embodiment of the present invention is divided for each component. The details of the text processing server 200 according to an embodiment of the present invention can be inferred from the description of the object name connection method and the text processing method according to the embodiment of the present invention described above, A detailed description will be omitted.

First, the text receiving unit 110 receives texts including one or more sentences from the terminals 10a and 10b.

Then, the polynomial calculation unit 120 calculates a polynomial word located in the text.

In addition, the neural network selecting unit 130 selects a neural network corresponding to the above-mentioned plural words among a plurality of neural networks.

Then, the neural network input unit 140 inputs all or part of words included in the text into the selected neural network.

Next, the meaning determination unit 150 determines the meaning of the ambiguous word in consideration of the output of the neural network.

Finally, the output unit 160 transmits the calculated output to the terminals 10a and 10b in consideration of the meaning of the ambiguous word.

Here, the plurality of neural networks is a neural network individually generated for each of a plurality of verbal words.

In addition, in the entity name concatenation system 100 according to an embodiment of the present invention, the plurality of neural networks may be a shallow neural network having a hidden layer within one layer.

The foregoing description is merely illustrative of the technical idea of the present invention, and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments described in the present invention are not intended to limit the technical spirit of the present invention but to illustrate the present invention. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents thereof should be construed as being included in the scope of the present invention.

10, 10a, 10b:
20: Communication network
30: Database
100: object name connection system
110:
120: Neural network prefix
130: Neural network input unit
140: entity name connection
200: Server
210:
220:
230: Neural network prefix
240: Neural network input unit
250:
260: Output section

Claims (13)

A method for linking an object name to a polymorphic word whose object name is located in a text containing one or more sentences,
Calculating a plural word in the text;
A neural network selecting step of selecting a neural network corresponding to the multiple words among a plurality of neural networks;
A neural network input step of inputting all or part of words included in the text into the selected neural network; And
And an entity name concatenation step of concatenating entity names of the ambiguous words in consideration of the output of the neural network,
Wherein the plurality of neural networks are generated individually for a plurality of plural words,
In the neural network selection step,
And selecting a neural network for the ambiguous word among the plurality of neural networks by using a list of polygons containing information on the neural network corresponding to each polygonal word. The method according to claim 1,
Wherein the plurality of neural networks are shallow neural networks in which no hidden layer exists. The method according to claim 1,
In the step of calculating the plural words,
And comparing the text with a predetermined list of plural words to compute a multiple word included in the text. delete The method according to claim 1,
And extracting at least one feature corresponding to the polyphonic word in the text,
Wherein, in the neural network input step, the extracted one or more qualities are input to the selected neural network. 6. The method of claim 5,
In the characteristic extraction step,
And extracting the qualities from words within a predetermined distance from the dominant word in the text. 6. The method of claim 5,
In the characteristic extraction step,
Extracting the qualities from words included in a sentence representing the text including the title, summary, and source of the text. 6. The method of claim 5,
In the characteristic extraction step,
And extracting the character by comparing the text with a predetermined list of qualities for each of the plural words. 6. The method of claim 5,
In the object name concatenation step,
Wherein the meaning of the ambiguous word included in the text is determined in consideration of at least one of a type, a number, and a weight of the qualities included in the text. A method of processing text transmitted from a terminal by a server,
A text receiving step in which a server receives text including one or more sentences from a terminal;
Calculating a plural word in the text;
A neural network selecting step of selecting a neural network corresponding to the multiple words among a plurality of neural networks;
A neural network input step of inputting all or part of words included in the text into the selected neural network;
A semantic determination step of determining the meaning of the ambiguous word in consideration of the output of the neural network; And
And an output transmission step of transmitting, to the terminal, an output calculated for the text in consideration of the meaning of the ambiguous word,
Wherein the plurality of neural networks are generated individually for a plurality of plural words,
In the neural network selection step,
Wherein the neural network is selected from among the plurality of neural networks using a plurality of neural network lists including information on neural networks corresponding to the respective neural networks. A computer program stored in a computer-readable medium for executing the steps of any one of claims 1 to 3 and 5 to 10 on a computer. A system for linking an object name to a multilingual word located in text containing one or more sentences,
A genetic term calculating unit for calculating a dominant word in the text;
A neural network selecting unit for selecting a neural network corresponding to the multiplex word among a plurality of neural networks;
A neural network input unit for inputting all or part of words included in the text by the selected neural network; And
And an object name connection unit for connecting the object name to the multilingual word in consideration of the output of the neural network,
Wherein the plurality of neural networks are generated individually for a plurality of plural words,
In the neural network pre-configuration,
Wherein the neural network is selected from among the plurality of neural networks by using a plurality of neural network lists including information on neural networks corresponding to the respective neural networks. A server for processing text transmitted from a terminal,
A text receiving unit for receiving text including one or more sentences from a terminal;
A polygonal word calculation unit for calculating a polygonal word located in the text;
A neural network selecting unit for selecting a neural network corresponding to the multiplex word among a plurality of neural networks;
A neural network input unit for inputting all or part of words included in the text by the selected neural network;
A meaning judging unit for judging the meaning of the ambiguous word in consideration of the output of the neural network; And
And an output unit for transmitting an output calculated for the text to the terminal in consideration of the meaning of the ambiguous word,
Wherein the plurality of neural networks are generated individually for a plurality of plural words,
In the neural network pre-configuration,
And selects a neural network for the ambiguous word among the plurality of neural networks by using a list of a plurality of terms including information on the neural network corresponding to each of the plurality of variants.

Images