KR20140056715A - An apparatus for opinion mining based on hierarchical categories and a method thereof - Google Patents

Abstract

The present invention relates to an apparatus for sentiment analysis based on hierarchical categories and a method therefore, including collecting analysis target documents corresponding to a selected category in order to extract sensibility data including keywords representing a target, attributes representing character or property of the keywords, expression representing evaluation for the attributes of the keywords and positive/negative values representing whether the expression is positive or negative and intensity of the expression according to a selected category from among a plurality of categories having a hierarchical tree structure; extracting keywords corresponding to the category for each sentence of the analysis target document; performing normalization that each sentence including the extracted keywords is reconstructed with a sentence including parts of speech to form the attributes and parts of speech to form the expression; and deriving the attributes from the reconstructed sentence and deriving expression corresponding to the derived attributes.

Description

[0001] APPARATUS FOR EMISSION ANALYSIS BASED ON HIERARCHICAL CATEGORY AND METHOD FOR THE SAME [0002]

The present invention relates to a technique for emotional analysis, and more particularly, to an apparatus and method for performing emotional analysis based on a hierarchical category.

Machine learning technology has been applied to various fields of natural language processing and its performance has been verified and applied to various real commercial systems. However, there is a problem that the time / monetary cost for constructing a learning set for machine learning, which is the most fatal weak point of the machine learning itself, is large, and the performance depends on the quality and amount of the learning set. In addition, compared with learning sets for languages such as English, which are actively under research in the field, language such as Korean is applied only to systems of technology research level because there is no built-in learning set that can be applied to actual commercial systems Level. Moreover, the research period for the emotion / opinion extraction field is relatively long, so the resources are not enough.

Knowledge / rule-based emotion / opinion extraction is a technique that has been adopted in many commercial emotion / opinion extraction systems so far. It can overcome the shortcoming of time / cost for building learning set for machine learning, It is relatively easy to control by knowledge and rules. However, it is impossible to regularize all kinds of emotions / expressions in natural language, and building and managing knowledge about all kinds of diversity also requires a lot of cost, and there is a limit to improving the accuracy of the system. Therefore, the existing emotion / opinion extraction system mainly focuses on emotion / opinion positive / negative classification of the entire document through the analysis of the expression part focused on the document.

It is an object of the present invention to provide an apparatus and method for performing emotional analysis using a category-based emotional knowledge system.

Another object of the present invention is to provide an apparatus and method for enhancing the accuracy of emotional analysis by performing preprocessing and normalization on each sentence of a document to be analyzed before emotional analysis is performed.

According to another aspect of the present invention, there is provided an apparatus for emotional analysis, the apparatus comprising: a plurality of categories having a hierarchical structure of a tree structure; An emotional data including an attribute indicating the nature or the nature of the main word, an expression indicating an evaluation of the attribute of the main word, and a positive / negative value indicating affirmation and denial and strength of the expression, A main word extracting unit for extracting main words corresponding to the selected category for each selected sentence, and normalizing each sentence including the extracted main word to reconstruct a sentence consisting of only parts of speech that can be attributes in each sentence and parts of speech that can be expressed, Derive an attribute from the sentence, and use the expression corresponding to the derived attribute And an emotional analysis unit for deriving emotional analysis.

The emotion analyzing unit reconstructs the expression of each sentence in the positive expression at the time of normalization, and extracts affirmation and negation and intensity of the expression and stores it as positive / negative value.

The emotion analyzing unit is characterized in that, when normalizing, in the case where there is a tubular expression having a form in which the attribute is forwardly modified by utilizing a tubular ending in each sentence, the emotion analyzing unit reconstructs the attribute into a general expression having a form of rearward modification.

The emotional analysis unit is characterized in that, at normalization, the relation between the attribute of the main clause and the expression of the subordinate clause is connected or disconnected according to the relationship between the main clause and the subordinate clause.

The emotional analysis unit judges whether or not there is an attribute corresponding to the expression in front of the expression, and when there is no attribute corresponding to the expression, the emotional analysis unit derives the attribute from the expression.

And the main word extracting unit extracts a main word of each sentence from the analysis target document by referring to a main word dictionary including main words corresponding to the selected category.

The main word extracting unit extracts a main word by referring to a grammar dictionary storing a plurality of sentence patterns having a comparative sentence in a case where there is a comparative sentence in the sentence.

The main word extracting unit estimates a main word in a previous sentence of one sentence when the main word can not be searched in any one sentence in the analysis target document and if the main word can not be estimated, And assigning one of the document subject terms related to the subject of the entire document as a subject word of one sentence.

And a preprocessor for performing preprocessing including sentence separation, spacing correction, morpheme analysis, and compound word recognition for each sentence of the analysis target document.

According to an aspect of the present invention, there is provided a method for emotional analysis, the method comprising: a plurality of categories having a hierarchical structure of a tree structure; An expression representing an evaluation of the attribute of the main word and a positive / negative value representing the affirmation and the negation and strength of the expression, and an analysis corresponding to the selected category Extracting subject words corresponding to the category for each sentence of the document to be analyzed, extracting part of speech that can be attributed in each sentence including the extracted subject word, A step of performing normalization that reconstructs a sentence composed of only parts of speech and a step of reconstructing And a sentence analysis unit for deriving an attribute from the sentence and deriving a representation corresponding to the derived attribute.

The step of performing the normalization is characterized by reconstructing the expression of each sentence as an affirmative expression, extracting affirmation and negation and intensity of the expression, and storing it as positive / negative values.

According to the present invention, it is possible to expect a sophisticated emotional analysis result according to the type and subject of a document by using the category classification system. It is possible to accurately grasp the degree of author's writing and emotion / opinion in connection with the object because of the technique of elaborately mapping the subject word which is the subject of emotion / opinion information. Algorithmization of the various phenomena of language which conventionally relied on a large amount of knowledge in the normalization process can improve the efficiency and convenience of knowledge management and enhance the performance of the system.

FIG. 1 is a diagram for explaining a tree structure of categories in an emotion dictionary according to an embodiment of the present invention.
2 is a diagram for explaining an apparatus for emotional analysis according to an embodiment of the present invention.
FIG. 3 is a diagram for explaining the detailed configuration of functions of the control unit 105 of the service apparatus 100 of FIG.
4 is a flowchart illustrating a method for emotional analysis according to an embodiment of the present invention.
FIG. 5 is a flowchart for explaining a method of extracting main words during emotional analysis according to an embodiment of the present invention.
6 is a flowchart illustrating a method for completing emotional data in emotional analysis according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description and the accompanying drawings, detailed description of well-known functions or constructions that may obscure the subject matter of the present invention will be omitted. It should be noted that the same constituent elements are denoted by the same reference numerals as possible throughout the drawings.

The terms and words used in the present specification and claims should not be construed as limited to ordinary or dictionary meanings and the inventor is not limited to the meaning of the terms in order to describe his invention in the best way. It should be interpreted as meaning and concept consistent with the technical idea of the present invention. Therefore, the embodiments described in the present specification and the configurations shown in the drawings are merely the most preferred embodiments of the present invention and are not intended to represent all of the technical ideas of the present invention, so that various equivalents It should be understood that water and variations may be present.

"Emotional analysis" according to the embodiment of the present invention recognizes the existence of emotion and opinion expression expressed by the creator of the text in a given text of the analysis target document, analyzes whether the corresponding expression is positive or negative, It is a function to detect the object of emotion / opinion expression in text and to provide a pair of emotion / opinion expression with corresponding object. For example, when the text "Samsung PAVV 7000 is good in design, but the picture quality is poor" is given, the emotion / opinion information of the author "the design is excellent" and "the image quality is low" Good ", and" poor quality "is a negative opinion. In addition, it finally judges that the sensitivity / opinion information is about "Samsung PAVV 7000" and outputs the result.

Based on the given examples, the elements extracted in the emotion analysis are defined as follows in the present invention.

For example, suppose that there is a sentence such as "Samsung PAVV 7000 is good in design, but picture quality is poor" in the document to be analyzed. "Design is excellent" (positive) The factors extracted from the emotional analysis are shown in Table 1 below.

Component Contents example Keyword
(Object) Subject of emotional analysis "Samsung PAVV 7000" property
(Feature) Target of Expression in Emotional Analysis "Design", "Quality" expression
(Expression) Affirmative or negative
Negative ratings or opinions "It's amazing," " Positive / negative value
(Polarity) Whether the expression is affirmative or not and its value Strong negative: -2
Weak negation: -1
Weak affirmation: 1
Strong affirmation: 2

As shown in Table 1, the elements extracted in the emotional analysis are the main word, attribute, expression, and positive / negative values, and these elements will be referred to as "emotional data" in the present invention. The main word is the object of emotional expression, and the attribute indicates the character, nature, and characteristics of the main word, and is the object of expression. Expression represents the evaluation of positive or negative opinion on the attributes of the main word. The positive / negative value is a value indicating whether the expression is affirmative or negative and its intensity. In Table 1, the positive / negative value is expressed as an integer between -2 and +2. Negative numbers indicate negative, positive numbers indicate positive, and magnitude of absolute value indicates positive / negative intensity. The intensity of positive / negative as used herein refers to the intensity of emotion / opinion expression. In the present invention, it is defined as distinguishing between "general emotion" and "strong emotion". "General emotion" is a form of general expression that expresses emotion / opinion and is a basic form such as "good", "decline", "good", "bad". "Strong emotions" refers to expressions that emphasize the emotion of positive / negative in the expression of positive / negative. The method for expressing "strong emotion" can be divided into two kinds, and the first is a method of expressing using degree / intensity adverbs. For example, "emotion / opinion information" I can express strong emotions by adding degree / intensity adverbs such as "I really like the design" or "I really like the design". In other ways, expressions themselves express strong emotions. For example, in expressing the feeling of "good", expressions such as "excellent" and "greatest" are expressions of strong emotion.

The apparatus for emotional analysis according to the embodiment of the present invention defines the emotional / opinion component as above and extracts a defined set of components. In the present invention, the knowledge of the emotional analysis system is managed by a dictionary, which is called an emotional dictionary. The emotion dictionary has positive / negative values corresponding to the attribute, expression, and attribute / expression pair defined above. These emotion dictionaries are constructed independently for each predetermined category. The construction of emotional dictionaries should be done by category because the set of attributes of the subject language group included in each category is different. For example, the attribute "shutter speed" is a commonly used attribute in the "camera" product set, but it does not exist in product sets such as "refrigerator", "diaper", and the like. Using these attributes without category identification will cause errors and management of emotional dictionaries will become inefficient, which will lead to performance degradation of emotional analysis system.

The category of the emotion dictionary is a tree structure and has a plurality of levels. FIG. 1 is a diagram for explaining a tree structure of categories in an emotion dictionary according to an embodiment of the present invention.

As shown, the category of the exemplary emotion dictionary is composed of a tree structure of 3-depth (3-depth). According to the embodiment of the present invention, the top category is called a domain, and has three domains of "person", "place", and "product". The structure of the categories below the domain is divided into the detailed classification according to the characteristics of the parent node (upper category), and the lowest category consists of the most detailed population in the domain. Figure 1 shows an example of an emotion category structure for the " product " domain.

Each category consists of entries of attributes and expressions independently, but there are also commonly used entries between similar categories. For example, assuming that "DSLR camera" and "camcorder" exist in the same level category, both the "DSLR camera" and "camcorder" categories can have a common property of "liquid crystal". In this case, if the attribute "liquid crystal" is managed for each category, the amount of emotional data becomes unnecessarily large, and the expressions for the attribute are also managed separately. In this case, according to the embodiment of the present invention, common emotion data belonging to each category is integrated and managed in the upper category. In actual emotion analysis, the emotion data of the upper category to which the category belongs can be inherited have.

In summary, the emotional analysis according to the embodiment of the present invention can be regarded as a process of registering emotional data including a main word, attribute, expression, and degree of a certainty for each category in a sentence dictionary classified into a category of a tree structure. An apparatus and method for such sensitivity analysis will be described below.

2 is a diagram for explaining an apparatus for emotional analysis according to an embodiment of the present invention.

Referring to FIG. 1, an apparatus for constructing emotional knowledge according to an embodiment of the present invention, that is, a service apparatus 100 may exist as an entity of a network connected to a network, It can communicate with entities. The service device 100 builds emotional knowledge by collecting data from at least one other entity present in the network. Typically, another entity may illustrate a web server.

To this end, the service apparatus 100 includes a communication unit 101, an input unit 102, a storage unit 103, a display unit 104, and a control unit 105.

The communication unit 101 is a device for connecting to a network and communicating with other entities. The communication unit 101 may further include a function of processing data according to the type of the network and the protocol applied to the network. For example, the communication unit 101 can communicate with a web server and receive an analysis target document such as a web document.

The input unit 103 is a means for receiving at least one of commands, selections, data, and information of a user of the service apparatus 100. The input unit 103 includes a plurality of input keys and function keys for inputting numeric or character information and setting various functions . The input unit 101 senses a key input of the user and transmits an input signal corresponding to the sensed key input to the controller 150. The input unit 101 may be an input device such as a keyboard, a keypad, a mouse, a joystick, and the like.

The storage unit 103 is an apparatus for storing data, and includes a main storage device and an auxiliary storage device. The storage unit 103 may store an operating system (OS), applications, and the like. The storage unit 103 may store each species data that the service apparatus 100 collects and generates. For example, the storage unit 103 may store a keyword dictionary containing main words corresponding to categories, an emotion dictionary storing attributes, expressions, positive / negative values, etc. corresponding to the categories, a grammar dictionary , And so on. Each kind of data stored in the storage unit 103 can be deleted, changed, or added according to a user's operation.

The display unit 104 is for displaying information generated according to use of the service apparatus 100 so that the user can recognize the information. Such information may include a document to be analyzed, a keyword, an attribute, an expression, a positive / negative value, a keyword, and the like.

The control unit 105 performs necessary operations to perform the method for building sensibility knowledge according to the embodiment of the present invention and may include a communication unit 101, an input unit 102, a storage unit 103, And the display unit 104 can be controlled. To this end, the control unit 105 may include a plurality of modules for realizing a method for building sensibility knowledge according to an embodiment of the present invention. Such a module may be composed of hardware or software.

The control unit 105 may be a processing apparatus that drives an operating system. For example, the control unit 105 may be a central processing unit (CPU). The control unit 105 moves the operating system from the auxiliary storage unit of the storage unit 103 to the main storage unit and then performs booting to drive the operating system. In addition, the control unit 105 can drive necessary applications necessary for performing emotional analysis according to an embodiment of the present invention.

FIG. 3 is a diagram for explaining the detailed configuration of functions of the control unit 105 of the service apparatus 100 of FIG.

Referring to FIG. 2, the control unit 105 includes a preprocessing unit 110, a main language extraction unit 120, and an emotion analysis unit 130.

The preprocessing unit 110 preprocesses the analysis target document before subject word extraction or before extraction of attributes, expressions, and the like. Such preprocessing includes sentence separation, spacing correction, morpheme analysis, compound word recognition, and the like, and the preprocessing unit 110 outputs a preprocessed sentence in units of sentences when the preprocessing is completed.

The main word extractor 120 extracts the main word for each of the preprocessed sentences. The main word extracting unit 120 may extract at least one of a title main word, a document main word, and a sentence main word as a main word.

The emotion analyzing unit 130 extracts attributes, expressions, and positive / negative values corresponding to the main word extracted by the main word extracting unit 120, and outputs emotional data including the main word, attribute, expression, and positive / negative values will be.

4 is a flowchart illustrating a method for emotional analysis according to an embodiment of the present invention.

The sensitivity analysis apparatus 100 collects the document to be analyzed in step S410. The collection of documents to be analyzed may be determined by the category in which the emotional analysis is to be performed. For example, if the category in which emotional analysis is to be performed is "camera ", the analysis target document can be collected from the corresponding website.

Next, in step S420, the emotion analyzing apparatus 100 extracts a main language word for each sentence in the analysis target document. At this time, the emotion analyzing apparatus 100 can extract the main word referring to the main word dictionary.

Then, in step S430, the emotional analysis apparatus 100 extracts attributes, expressions, and positive / negative values corresponding to the extracted main word to derive emotional data.

Hereinafter, a description will be given of a step of extracting the above-mentioned main words and a step of extracting emotional data by extracting attributes and expressions.

FIG. 5 is a flowchart for explaining a method of extracting a main word in emotional analysis according to an embodiment of the present invention, and explains the step S420 described above in more detail.

The main words are managed by the main dictionary, and the main dictionary is managed according to the predetermined category classification system like the emotional dictionary. As in the embodiment of the present invention, if a dictionary for general object names is managed for each subdivided category without managing all the entries collectively, it is possible to perform a sophisticated analysis on the documents corresponding to the performance categories among the subject words.

Keyword extraction basically has the following assumptions.

1. The emotional / opinional document must have the subject language to which it relates.

2. A document with emotion / opinion may have multiple keywords.

3. In the document, subject words or subject words have areas that they control.

Since emotional analysis provides opinions, reputation, etc. on the subject, it is essential that the first assumption is made. Otherwise, the emotional analysis can not be provided. In particular, in the case of the assumption No. 3, in the present invention, only the main word is extracted, and the extracted main word is recognized to the region governed by the document, and when the emotion expression is extracted later, . When extracting the main words, we have the following three types of main word extraction types.

1. Title Keyword: The subject (s) extracted from the title of the document.

2. Document Keyword: Among the main words extracted from the body of the document, the most frequently expressed keywords

3. Sentence Key words: Key words extracted from each sentence of the document.

In the document, subject match is given to the subject dictionary with the exact match (Exact Match) that finds the exact match of the word and the subject word held in the dictionary, and if it fails, a flexible matching Relaxed Match. Basically, since the subject dictionary has most variations on the subject word, it reduces the confidence value for the extracted subject word when attempting a flexible matching (Relaxed Match).

In Fig. 5, the derivation of the main word is performed on a sentence unit basis. The main word search is searched to see whether any of the words described in the main word dictionary in the sentence dictionary matches with the main word dictionary, and when the extraction is completed, Information can be added. To this end, the service apparatus 100 first derives a title main word and a document main word for an analysis target document, and then performs the following procedure for each sentence of the analysis target document.

If one sentence is input in step S510, the service apparatus 100 determines whether a main word exists in the sentence input in step S515. If it is determined in step S515 that the main word exists in the inputted sentence, the process proceeds to step S520, and if the main word does not exist in the inputted sentence, the process proceeds to step S535.

In step S520, the service device 100 determines whether a comparison (comparison) is included in the sentence. At this time, if there is a comparison in the sentence, the service apparatus 100 proceeds to step S525, and if not, proceeds to step S530. The problem with sentence subject word extraction is the comparison. In simple sentences, when a subject word is extracted, the subject word dominates the sentence in most cases. For example, in the sentence "Canon 600D has a faster shutter speed", a product called "Canon 600D" dominates the entire sentence, and the emotion expression extracted from that sentence, "Shutter speed fast", refers to "Canon 600D" This is clear. However, the sentence "The Canon 600D has a slower shutter speed than the Nikon D70" has two objects, the "Canon 600D" and the "Nikon D70", and moreover, the "Nikon D70" It is closer. However, the actual "shutter speed slow" sensibility must be assigned to the subject "Canon 600D". Accordingly, the service apparatus 100 manages the plurality of sentence patterns having the comparative sentence as a grammar dictionary for storing the sentence patterns, and finally assigns proper keywords to expressions related to the comparison used in the sentence. Accordingly, in step S525, the service apparatus 100 extracts the keyword using the grammar dictionary and the keyword dictionary in the corresponding sentence, considering the comparison related expressions. On the other hand, in step S530, the service device 100 extracts the main word of the sentence using the main word dictionary in the corresponding sentence.

If the subject sentence does not exist in the input sentence, the service apparatus 100 determines whether the main word can be estimated in step S535. That is, since all the sentences do not include the subject word, when the subject word is not extracted in the sentence, the service apparatus 100 must estimate the subject word.

Accordingly, if it is determined in step S535 that the main language estimation is possible, the service device 100 proceeds to step S540 so as to extract the main word estimated by the main word estimation, as the main word. In order to estimate the main word, first, it searches the main word found in the near. In the embodiment of the present invention, the main word is searched by searching up to the previous three sentences of the inputted sentence.

On the other hand, if it is determined in step S535 that the subject term can not be estimated, that is, if the subject term can not be extracted in the sentence according to the above description, or the subject term can not be estimated through the surrounding sentence, Should be assigned.

Accordingly, the service apparatus 100 determines whether the document main language is the main language or the title main language is the main language in step S545. The criterion first places the document subject word as the highest priority. To extract the document subject words, the most frequently extracted subject words are selected from the subject words from the document to the sentences in which the subject word extraction is performed. However, if the most frequently extracted main word among multiple subject words does not exceed 50% of the extracted main words, the main word of the document is not selected as the main word. This is because the shorter the number of short sentences, the more likely it is to find the main word in the title of the document and the higher the probability that the main word (including the title) will be extracted at the beginning of the document. In this case, if there is a subject word extracted from the title, it has a higher priority than the document subject word. The extraction of the title main word takes the same way as the extraction of the sentence main word, and if there are differences, it has lower reliability than the sentence main word.

Accordingly, the service apparatus 100 determines whether to use the document main language as a main language or the title main language as a main language in step S545 according to the above-described criteria, extracts the document main language as a main language in step S550, In step S560, the title main word is extracted as a main word.

FIG. 6 is a flow chart for explaining a method for completing emotional data during emotional analysis according to an embodiment of the present invention, and explains the step S430 described above in more detail.

As described above, when all the subject words are extracted from the analysis target document, the service apparatus 100 performs processing on a sentence unit basis to complete the emotion data corresponding to the extracted main word. At this time, the service apparatus 100 refers to the emotion dictionary, and the emotion dictionary is loaded according to the category to which the analysis subject document belongs, and analyzes each sentence with knowledge of the attributes and expressions.

First, in step S610, the preprocessing unit 110 performs preprocessing on the document to be analyzed. That is, before the actual emotional analysis is entered, the preprocessing unit 110 performs preprocessing including sentence separation, spacing correction, morphological analysis, compound word recognition, and the like for each sentence of the analysis target document.

When the preprocessing process is completed, the service apparatus 100 normalizes each sentence in step S620, and derives a keyword for each sentence. In this case, the normalization reconstructs the string as it is expressed by the creator of the document to be analyzed as the normalized expression as much as possible, removes unnecessary information, facilitates the advance search, and adds clue information as clues for analysis during reconstruction , And to utilize the emotional dictionary which is possessed efficiently.

That is, the normalization refers to a task of generalizing each sentence to be analyzed as a form of only attributes and expressions, and constituting the remaining information as clue information. This normalization basically includes 1) reconstitution into a sentence consisting of only morpheme necessary to derive emotional data, 2) selectively processing the relationship between attributes and expressions in a sentence, 3) processing the expression of a tubular type, 4) Lt; RTI ID = 0.0 > and / or < / RTI >

1) Reconstructing sentences composed of necessary morphemes to derive emotional data

The service apparatus 100 reconstructs the sentences that have been subjected to the preprocessing such as morpheme analysis and compound word recognition (S610) by only morphemes and compound words required for emotional analysis. The reconstructed sentence is reconstructed into a restored form including a general noun which can be used as an attribute, a genitive term such as a proper noun, an adjective which can be used as a cognitive phrase and an expression, a verb such as a verb, and a verb phrase. Other parts of speech are used as clue information in extracting attributes and expressions and are not included in the reconstructed sentence. Thus, the reconstructed sentence is defined as a key term.

2) Handling the relationship between attributes and expressions in sentences

The emotional analysis unit 130 sets the relation between the attribute of the main clause and the expression of the subordinate clause to be connected or disconnected according to the relationship between the main clause and the dependent clause. There is a complex relationship between attributes and expressions in actual sentences. For example, you can extract two expressions of emotion: "Design is great" and "Size is burdensome" in the sentence "The design is great, but the size is burdensome". Here, the attribute "design" is connected only to the expression "good", and the "burden" part is disconnected. In this case, the emotional analysis unit 130 stores the connection term "but" as the clue information, recognizes that the sentence can be separated into two clauses by utilizing it, and analyzes whether the following clause is a subordinate clause or a COP And finally ends the relationship between "design" and "burden". If the following clause is in the form of a subordinate clause, the emotional analysis unit 130 does not break the relationship between "design" and "burden". For example, in a sentence such as "Design is good and beautiful", "design" has a relationship with both "good" and "beautiful" expressions.

3) Expression processing of tubular type

The sensitivity analysis unit 130 reconstructs the tubular expression as a general expression. If it is reconstructed as a regular expression, the sentence can be a keyterm. In the embodiment of the present invention, a tubular expression refers to a form in which a target attribute is modified from the front by utilizing a tubular ending. For example, "good design" is good expression of the attribute design, while the rear form is the general form expression, "good design" is defined as a tubular expression. The main feature of the tubular expression is that there is an attribute that receives the expression, and the attribute and the section of the expression exist independently in the sentence. In addition, all tubular expressions can be reconstructed as regular expressions. The following Table 2 is intended to illustrate an example of normalization that reconstructs this tubular representation as a generic representation.

In this way, in the normalization process, the emotion analyzer 130 basically converts the tubular description into the general description to generate the keyword. Then, the matching area of the attribute and the expression is limited to the tubular expression.

4) positive / negative value processing;

The emotion analyzing unit 130 reconstructs positive and negative expressions as affirmative expressions to generate a keyword, and stores affirmative and negative expressions and intensities as positive / negative values.

According to the embodiment of the present invention, in the process of extracting the kittens according to the normalization, the negative expression processing of the expression and the intensity recognition of the expression are also performed. First, when we look at simple negative expressions, we do not add negative expressions to emotion dictionaries basically. It is very unnecessary and inefficient to have all kinds of expressions such as negative expressions for each expression. Therefore, the emotional dictionary has all the knowledge in the form of the most simplified affirmative sentence (affirmative expression), but recognizes the negative expressions in the actual sentence and adjusts the positive / negative value. For example, assuming that the sensibility knowledge of "good design" has a value of "general affirmative" or "1" as the basic positive / negative value, the actual "bad design" And the negative expression and the intensity are stored by being mapped to " general negative ", that is, " -1 " through positive / negative values.

In the present invention, in addition to the basic negative expressive processing, the emotional intensity and positive / negative reversal appearing in the combination of multiple negation, degree adverb and negation are also handled. Multiple negation refers to a phenomenon in which negation becomes negative again and becomes affirmative, and it is a sentence such as "design is not bad". Negative words appear in the sentence, but these negative words cancel each other and become positive expressions again. In addition, this type of negation is combined with the degree adverb to deal with the change of polarity and intensity. As described above, the emotional expression is defined to be able to express strong emotion by degree adverb. However, the interpretation of "the design is very good" and "the design is not very good" can be changed. The first sentence becomes "strong positive" of "good design", but the second sentence can be interpreted as "weak positive", that is, "design is slightly better". In addition, "design is not very bad" in this sentence, "very" except for adverbs, except interpretation, "good design" is interpreted, but "very" including the interpretation, "the design is a little bad "Or" weak denial ". In all of these cases, the emotion analyzing unit 130 reconstructs using the "design is good" and the positive / negative values, and the positive / negative value is the clue information.

When normalization is completed as described above, the service apparatus 100 performs "aspect analysis" in step S630, thereby filtering sentences that can not be subjected to emotional analysis.

&Quot; Analysis of Modes " defined in the present invention is carried out. In the embodiment of the present invention, the mode refers to whether the analysis target sentence, clause, or the like is assumed, expected, compared, questioned, or exempted from the analysis target, and the extracted keyword is reconfigured according to the type. For example, "It would be nice if the design improved ... Quot; is recognized in the system as a pattern of the " expectation " aspect, and the extracted keystore, " improved design " The information on this aspect is patterned and managed in a manner analysis dictionary.

Next, in step S640, the emotional analysis unit 130 searches for sentences and extracts attributes and expressions for each sentence, using emotion dictionaries and kits. That is, the emotional analysis unit 130 first extracts matched attributes among the attributes registered in the emotion dictionary of the category in the sentence. Then, the emotional analysis unit 130 extracts matched matching expressions among expressions of attributes extracted from the emotion dictionary of the category. In addition, the emotion analyzing unit 130 reflects the reliability value according to the accuracy of the match of all searches. In the process of extracting the kittens, unnecessary information is removed, and cue information is also stored separately. Therefore, if there is a difference between the attribute and the expression over a certain distance, there is a high possibility that there is a noise to break the relationship between them. Accordingly, when extracting a representation of an attribute, the emotion analyzing unit 130 does not extract the expression of the attribute when the attribute and expression are greater than a predetermined distance. For example, you can have a distance constraint of up to five tokens in a match of attribute and expression.

On the other hand, in the case of a form in which there is no attribute and extracting only attributes and expressions, it is defined as " null attribute expression " in the present invention.

In the case of the " NULL attribute expression ", the emotion analysis unit 130 determines whether an attribute corresponding to the expression exists before the expression, and if there is no attribute corresponding to the expression, the attribute can be derived from the expression have.

Basically, in expressing emotion or opinion, it is common to make a specific attribute referring to an object, but there is a general evaluation of the object or an expression in which the attribute is implicitly expressed without mentioning the attribute. For example, in a sentence such as "FULLY GOOD", the attribute does not exist, but it expresses the opinion on the specific object and includes the comment such as the general comment on the object. Also, the expression "expensive" such as "this product is expensive" implicitly includes a property called "price", but does not refer to the property on the surface. These entries are managed as " null attribute representations " in emotion knowledge. The emotional analysis unit 130 performs the search only when the attribute search fails in the emotional analysis. The search is the same as a regular expression search, but has constraints. The extraction is performed only when the search is successful and the target of the attribute or expression is not completely present in front of the expression. For example, in the camera review document, "good" in the phrase "the weather is fairly good" refers to "weather", but since the "weather" does not exist in the sensitivity category knowledge of the camera category, Will be a successful search.

Next, in step S650, the emotional analysis unit 130 outputs emotional data including a main word, attribute, expression, and positive / negative values for the sentence from which the attribute and expression are extracted. In this case, the positive / negative value may be a positive / negative value derived from the normalization.

The method for emotional analysis according to the present invention may be implemented in a form of software readable by various computer means and recorded in a computer-readable recording medium. Here, the recording medium may include program commands, data files, data structures, and the like, alone or in combination. Program instructions to be recorded on a recording medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. For example, the recording medium may be an optical recording medium such as a magnetic medium such as a hard disk, a floppy disk and a magnetic tape, a compact disk read only memory (CD-ROM), a digital video disk (DVD) Includes a hardware device that is specially configured to store and execute program instructions such as a magneto-optical medium such as a floppy disk and a ROM, a random access memory (RAM), a flash memory, do. Examples of program instructions may include machine language code such as those generated by a compiler, as well as high-level language code that may be executed by a computer using an interpreter or the like. Such hardware devices may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, It will be apparent to those skilled in the art. Furthermore, although specific terms are used in this specification and the drawings, they are used in a generic sense only to facilitate the description of the invention and to facilitate understanding of the invention, and are not intended to limit the scope of the invention.

The present invention relates to an apparatus and method for emotion analysis based on a hierarchical category, and a method for the same. The present invention relates to an apparatus and method for emotion analysis based on a hierarchical category, wherein a plurality of categories have a hierarchical structure of a tree structure, To extract the emotion data including the main word indicating the nature or the nature of the main word, the expression indicating the evaluation of the attribute of the main word, and the positive / negative value indicating the affirmation and the negation and the strength of the expression, A main word extracting unit for extracting main words corresponding to the selected category for each sentence from the sentence, and normalizing each sentence including the extracted main word to reconstruct a sentence consisting of only parts of speech that can be attributed in each sentence and parts of speech that can be expressed , Derives the attribute from the reconstructed sentence, An emotional analysis unit for deriving a corresponding expression, and an apparatus for emotional analysis. According to the present invention, it is possible to expect a sophisticated emotional analysis result according to the type and subject of a document by using a category classification system. It is possible to accurately grasp the degree of author's writing and emotion / opinion in connection with the object because of the technique of elaborately mapping the subject word which is the subject of emotion / opinion information. Algorithmization of the various phenomena of language which conventionally relied on a large amount of knowledge in the normalization process can improve the efficiency and convenience of knowledge management and enhance the performance of the system. This is not only a possibility of commercialization or sales, but also a possibility of being industrially applicable since it is practically possible to carry out clearly.

100: service apparatus 101: communication unit
102: input unit 103: storage unit
104: Display unit 105:
110: preprocessing unit 120: main word extracting unit
130: Emotion analysis section

Claims (12)

A plurality of categories having a hierarchical structure of a tree structure, and each of the plurality of categories has a hierarchical structure of a plurality of categories, wherein the keywords include a main word indicating an object, an attribute indicating a nature or a property of the main word, A main word extracting unit for extracting emotional data including positive / negative values indicating affirmation and denial of expressions and strengths, and subject terms corresponding to the selected category for each sentence from an analysis target document;
Wherein each sentence including the extracted main word is normalized to reconstruct a phrase consisting of part-of-speech that can be the attribute and parts-of-speech that can be the expression in each sentence, deriving the attribute from the reconstructed sentence, And an emotion analyzing unit for deriving a representation corresponding to the attribute. The method according to claim 1,
The emotional analysis unit
Wherein the expression of each sentence is reconstructed as an affirmative expression at the time of normalization, and the positive and negative states and strengths of the expression are extracted and stored as a positive / negative value. The method according to claim 1,
The emotional analysis unit
Wherein, in the normalization, when there is a tubular expression having a form of modifying an attribute in front by using a tubular ending in each sentence, the attribute is rearranged into a general expression having a form of rearward modifying . The method according to claim 1,
The emotional analysis unit
Wherein the relationship between the attribute of the main clause and the expression of the subordinate clause is set to be connected or disconnected according to the relation between the main clause and the subordinate clause at the time of the normalization. The method according to claim 1,
The emotional analysis unit
Determining whether or not an attribute corresponding to the expression exists in front of the expression, and deriving an attribute from the expression when the attribute corresponding to the expression does not exist. The method according to claim 1,
The main word extracting unit
Wherein the main word of each sentence is extracted from the analysis target document by referring to a main word dictionary including main words corresponding to the selected category. The method according to claim 1,
The main word extracting unit
Wherein a grammar dictionary storing a plurality of sentence patterns having a comparative sentence is referred to when a sentence has a comparative sentence, and the main word is extracted. The method according to claim 1,
The main word extracting unit
If the subject term can not be searched in any sentence in the analysis target document,
Estimating a subject word in a previous sentence of the one sentence,
If the subject word can not be estimated,
Wherein one of the title key words related to the title of the document to be analyzed and the document key words related to the subject of the entire document is assigned as a key word of the one sentence. The method according to claim 1,
Further comprising a preprocessor for performing preprocessing including at least one of sentence separation, spacing correction, morpheme analysis, and compound word recognition for each sentence of the analysis target document. A plurality of categories having a hierarchical structure of a tree structure, and each of the plurality of categories has a hierarchical structure of a plurality of categories, wherein the keywords include a main word indicating an object, an attribute indicating a nature or a property of the main word, Collecting analysis target documents corresponding to the selected category to extract emotional data including positive / negative values indicating affirmation and denial of expression and intensity;
Extracting subject terms corresponding to the category for each sentence of the analysis target document;
Performing normalization in each sentence including the extracted main word to reconstruct the sentence consisting of only parts of speech that can be the attribute and part of speech that can be the expression; And
Deriving the attribute from the reconstructed sentence and deriving a representation corresponding to the derived attribute;
The method comprising the steps of: 11. The method of claim 10,
The step of performing the normalization comprises:
Wherein the expression of each sentence is reconstructed as an affirmative expression, and positive and negative indications and strengths of the expressions are extracted and stored as positive / negative values. A computer-readable recording medium recording a program for performing the method for emotional analysis according to claim 10 or 11.

Images