WO2014208880A1

WO2014208880A1 - Word comfort/discomfort index prediction apparatus and method therefor

Info

Publication number: WO2014208880A1
Application number: PCT/KR2014/003216
Authority: WO
Inventors: 이수원; 이강복
Original assignee: 숭실대학교산학협력단
Priority date: 2013-06-26
Filing date: 2014-04-15
Publication date: 2014-12-31

Abstract

An apparatus for predicting a comfort/discomfort index of a word and a method therefor are disclosed. The disclosed apparatus for predicting a comfort/discomfort index of a word calculates an emotional association between a word and a plurality of comparative words respectively, and calculates an emotional association between at least one pre-defined reference word and the plurality of comparative words respectively. The apparatus comprises a calculation unit for calculating an emotional similarity between the word and the at least one reference word respectively; and a prediction unit for predicting a comfort/discomfort index of the word using the at least one emotional similarity and the pre-defined comfort/discomfort index of the at least one reference word.

Description

Apparatus and method for predicting word pleasure-discomfort index

Embodiments of the present invention relate to an apparatus and method for predicting a pleasantness-discomfort index of words, and more particularly, to further refine text emotions by predicting the degree of emotion (comfort-discomfort index) of words which has not been considered in word emotion prediction techniques. An apparatus and method for enabling analytical techniques.

Techniques for predicting emotion of words include technology using word association, technology using large-scale dictionary, and technology using rating of product reviews.

In this related art, "automatic construction of a feature-specific predicate affirmation / negative dictionary for classification of product review polarity (Song Jong Suk, Lee Soo Won, Journal of KIISE: Software and Applications, Vol. 38, No. 3 (2011.3), p157 ~ p168) is known. .

On the other hand, the prior art has a disadvantage that can not conduct a multi-sensitivity analysis because it analyzes the words in two emotions of positive-negative, and can not classify 'strong positive' and 'weak positive' is impossible advanced analysis.

In order to solve the problems of the prior art as described above, in the present invention, by predicting the degree of emotion (comfort-discomfort index) of the words that are not considered in the word emotion prediction technology, the more advanced text emotion analysis technology is enabled. An apparatus and method for predicting annoyance index are proposed.

Other objects of the present invention may be derived by those skilled in the art through the following examples.

According to a preferred embodiment of the present invention to achieve the above object, in the apparatus for predicting the comfort-discomfort index of a word, calculating the emotional association of each of the word and one or more comparison words, at least one preset A calculation unit configured to calculate an emotion correlation between each of the reference words and each of the one or more comparison words, and calculate an emotion similarity between the word and each of the at least one reference word; And a predicting unit predicting the pleasure-discomfort index of the word based on the at least one emotional similarity and the preset pleasure-discomfort index of the at least one reference word. A prediction apparatus is provided.

The calculator uses a ratio of a probability that the word or the reference word and the comparison word appear independently in one paragraph, and a probability that the word or the reference word and the comparison word appear together in one paragraph. The emotional correlation between the word or the reference word and the comparison word may be calculated.

The calculation unit may be configured to calculate a first vector including, as an element, an emotional correlation between each of the words and the one or more comparison words, and a second vector including, as an element, an emotional relation between each of the basic word and the one or more comparison words, An emotional similarity between the word and the reference word may be calculated using the angle between the first vector and the second vector.

The prediction unit may predict the pleasant-discomfort index of the word using the following equation.

Here, P / UNPp is the pleasant-to-off index of the predicted word, SUM is the sum of the emotional similarity of the word and the at least one reference word, n is the index of the reference word, N is the number of at least one reference word , FSn denotes the nth emotional similarity, and P / UNPn denotes a preset pleasant-discomfort index of the nth reference word, respectively.

According to another embodiment of the present invention, there is provided a method of predicting a pleasant-discomfort index of a word, the method comprising: calculating an emotional association of each word with one or more comparison words; Calculating an emotional correlation of each of at least one preset reference word and each of the one or more comparison words; Calculating an emotional similarity of each of the word and the at least one reference word; Predicting the pleasant-discomfort index of the word using the at least one emotional similarity and a preset pleasant-discomfort index of the at least one reference word; Provided is a method for predicting a pleasant-discomfort index of a word comprising a.

According to the present invention, a more advanced text emotion analysis technique is enabled by predicting the degree of emotion (comfort-discomfort index) of words not considered in the word emotion prediction technique.

1 is a diagram illustrating a schematic configuration of a word comfort-discomfort index prediction system according to an embodiment of the present invention.

2 is a diagram illustrating a schematic configuration of a reference word processor according to an exemplary embodiment of the present invention.

3 is a diagram illustrating a schematic configuration of a word preprocessor according to an exemplary embodiment of the present invention.

FIG. 4 is a diagram illustrating a schematic configuration of an apparatus for predicting word pleasantness / discomfort index according to an embodiment of the present invention.

5 is a flowchart illustrating a schematic process of a method of predicting a word comfort-discomfort index according to an embodiment of the present invention.

As the invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing the drawings, similar reference numerals are used for similar elements.

In the present invention, in order to predict the pleasure-discomfort index of the word (the degree of emotion), if the association with words appearing simultaneously in one sentence is similar, the emotion of two emotional words (that is, words for expressing the emotion of a person) The degree will be similar. ' To do this, we use a list of some emotional words and their associated pleasure-discomfort index to calculate the association between the words and the words that appear simultaneously. After calculating the association, the similarity between the emotional words is obtained, and the similarity is used to predict the pleasure-discomfort index of the word.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Referring to FIG. 1, the system for predicting a pleasantness-discomfort index 100 of a word according to an exemplary embodiment of the present invention includes a reference word processor 110, a word preprocessor 120, and a word comfort-discomfort index predictor 130. It includes. Hereinafter, the function of each component will be described in detail with reference to FIG. 1. In addition, for convenience of description, a word for predicting a Pleasant-Unpleasant will be referred to as a 'target word'.

The reference word processor 110 preprocesses a word related to at least one emotion used to predict the pleasure-discomfort index of the 'target word'. Hereinafter, for convenience of description, the word related to the emotion is referred to as a 'reference word' as a meaning distinguished from the 'target word' for predicting the comfort-discomfort index.

In more detail, the reference word processor 110 presets the comfort-discomfort index for at least one 'reference word' and 'reference word'. To this end, the reference word processor 110 may include a reference word module 111, a morpheme analyzer 112, and a stopword removal module 113 as shown in FIG. 2. Here, the morpheme analysis unit 112 performs a task of tagging parts of speech appropriate to each word (eg, a task of informing that 'love' is a 'noun'), and the stopword module 113 is loved. This function removes parts of the word that have no real meaning, such as do.

Next, the word preprocessor 120 extracts a preprocessed sentence list from sentences such as online content, and extracts preprocessed words therefrom. To this end, the word preprocessor 120 includes a crawler 121, a preprocessor 122, a morpheme analyzer 123, an automatic string spacing module 124, and some colloquial characters. The speech correction module 125, the sentence classification module unit 126, and the stop word removal module unit 127 are included.

Subsequently, the word comfort-discomfort index predictor 130 predicts the comfort-discomfort index for the preprocessed word.

In more detail, as shown in FIG. 4, the word comfort-discomfort index prediction device 130 includes a calculator 131 and a predictor 132. 5 is a flowchart illustrating the overall flow of the operation of the word comfort-discomfort index prediction apparatus 130 (the word comfort-discomfort index prediction method) according to an embodiment of the present invention. Hereinafter, functions of each component and operations performed by each step will be described in detail with reference to FIGS. 4 and 5.

In operation S510, the calculator 131 calculates an emotional correlation between the 'target word' and the 'comparative word'. In addition, the calculation unit 131 iteratively calculates the emotional correlation with respect to the entirety of one or more 'comparative words'.

For example, if you have a "target word" called "exam" and a "comparative word" such as "result" and "thrill", there is an emotional relationship between "exam" and "result", There is also an emotional connection between "a breath". Therefore, the calculation unit 131 probabilistically calculates how emotionally the relationship between the "test (target word)" and the "test (comparative word 1)" and the "thirth (comparative word 2)" is obtained.

According to an embodiment of the present invention, the calculation unit 131 is a probability that the 'target word' and the 'comparative word' appear independently in one paragraph, and the 'target word' and the 'comparative word' appear together in one paragraph. The emotional correlation between the 'target word' and the 'comparative word' can be calculated by using a ratio of the probability to do. As described above, this is based on the premise that "if the associations with words appearing in one sentence are similar, the two emotion words will have similar emotions."

As an example, the calculation unit 131 of the present invention may calculate the emotional association between the 'target word' and the 'comparative word' using a Pointwise Mutual Information (PMI) index as shown in Equation 1 below.

Equation 1

Where A is the 'target word', B is the 'comparison word', PMI (A, B) is the emotional association between the 'target word' and the 'comparison word', and P (A) × P (B) is the 'target word' And the probability that 'comparative words' appear independently in one paragraph, and P (A∩B) means the probability that 'target word' and 'comparative words' appear together in one paragraph.

In operation S510, the calculator 131 further calculates a first vector having an emotional relationship between each of the 'target word' and the one or more 'comparative words'. For example, emotional linkage (first vector) and 'test' (target word) and 'sorry, regret' for 'Burn' (target word) and 'enjoy, entertain, fun, sad, joy' (hereinafter, comparison word) , Disregard, self-reproach, and breathtaking '(hereinafter referred to as comparison word) have the emotional relations (first vector) as shown in Table 1.

Table 1

	Enjoy (B1)	Heung (B2)	Fun (B3)	Regrets (B4)	Happy (B5)
Gold (A)	[6.04, 5.53, 4.67, 4.60, 4.01]-> first vector
	Sorry (B1)	Regret (B2)	Ignore (B3)	Own self (B4)	Breathtaking (B5)
Test (A)	[5.87, 4.87, 4.87, 4.59, 3.92]-> first vector

Next, in step S520, the calculator 131 calculates the emotional association between the preset 'reference word' and the 'comparative word', similarly to the 'target word', but repeatedly for one or more 'comparative words'. To calculate the emotional association.

According to the exemplary embodiment of the present invention, similarly to the above description, the calculator 131 may determine the probability that the 'reference word' and the 'comparative word' appear independently in one paragraph, and the 'reference word' and the 'comparative word'. The ratio of the probability of appearing together in a paragraph can be used to calculate the emotional association between 'reference word' and 'comparative word'. As an example, the calculation unit 131 of the present invention may calculate the emotional correlation between the 'reference word' and the 'comparative word' using the PMI index as shown in Equation 2 below.

Equation 2

Where C is the reference word, B is the comparison word, PMI (C, B) is the emotional relationship between the reference word and the comparison word, and P (C) × P (B) is the reference word And the probability that 'comparative words' appear independently in one paragraph, and P (C∩B) means the probability that 'reference word' and 'comparative words' appear together in one paragraph.

In addition, the calculator 131 calculates a second vector having an element associative relationship between each of the 'reference word' and the one or more 'comparative words', and the calculator 131 respectively calculates at least one 'reference word'. Calculate a second vector (ie, at least one second vector) with respect to, in other words, the calculation unit 131 calculates an emotional correlation between each of at least one preset reference word and one or more comparison words. As a result, at least one second vector is calculated.

In operation S530, the calculator 131 calculates an emotional similarity of each of the 'target word' and the at least one 'reference word'. Emotional similarity is a measure of how emotionally relevant is by comparing 'target word' with 'reference word'.

According to an embodiment of the present invention, the calculating unit 131 may calculate the emotional similarity between the 'target word' and the 'reference word' using the angles of the first vector and the second vector, which is represented by the following equation. It can be expressed as 3.

Equation 3

here,

Is the cosine similarity, and θ means the angle between the first vector and the second vector. Therefore, the smaller the angle between the first vector and the second vector, that is, the greater the cosine value, the greater the emotional similarity between the 'target word' and the 'reference word'.

Meanwhile, a plurality of 'reference words' exist, and the number of at least one 'reference words' may be higher k (an integer of 1 or more) having high emotional similarity among the plurality of 'reference words'.

Table 2 below shows an example of cosine similarity when 'target word' is 'happy', five 'reference words' are 'love, good, happy, happy, sad' and 'target word' is 'own self' Is a table showing an example of the cosine similarity when the five 'standard words' are "regret, anxiety, trouble, trouble, sin."

TABLE 2

	Love (C1)	Good (C2)	Joy (C3)	I am glad (C4)	Sad (C5)
Happiness (A)	0.35	0.29	0.27	0.24	0.21
	Regrets (C1)	Worry (C2)	Chong (C3)	Worry (C4)	Sin (C5)
Own self (A)	0.15	0.12	0.10	0.10	0.09

Next, in step S540, the prediction unit 132 predicts the pleasant-discomfort index of the word by using the at least one emotional similarity and the predetermined pleasant-discomfort index of the at least one reference word. Here, the predetermined pleasant-discomfort index of the at least one reference word may be set in the reference word processor 110 as described above.

In addition, according to an embodiment of the present invention, the prediction unit 132 may predict the pleasure-discomfort index of the word by using Equation 4 below.

Equation 4

Where P / UNPp is the predicted pleasure-discomfort index of the word, SUM is the sum of emotional similarities of the 'target word' and at least one 'reference word', n is the index of the 'reference word', and N is at least one ' The number of reference words', FSn denotes the emotional similarity between the word and the nth reference word, and P / UNPn denotes the preset comfort-discomfort index of the nth 'reference word'.

As an example, the 'target word' is 'response', the five 'reference words' are 'regret, anxiety, sticky, trouble, sin', and the five pleasure criteria of the 'reference word' are set in order. In the case of "2.45, 2.94, 2.57, 2.75, 2.20", and the similarity of each of the "target word" and the "reference word" calculated by the calculating unit 131 is "0.155, 0.127, 0.107, 0.106, 0.099", The sum of the similarities (SUM) is 0.594, and the predicted discomfort index (P / UNPp) of the word may be 2.588 as shown in Equation 5 below.

Equation 5

It can be seen that the reference word processing unit 110 is similar to the actual value "2.24" calculated according to a technique known in "list creation and dimension search of Korean test words" for "target word (response)".

As described above, according to the present invention, there is an advantage of enabling a more advanced text emotion analysis technique by predicting the degree of emotion (comfort-discomfort index) of words not considered in the word emotion prediction technique.

In addition, embodiments of the present invention can be implemented in the form of program instructions that can be executed by various computer means may be recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the media may be those specially designed and constructed for the purposes of the present invention, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Examples of program instructions such as magneto-optical, ROM, RAM, flash memory, etc. may be executed by a computer using an interpreter as well as machine code such as produced by a compiler. Contains high-level language codes. The hardware device described above may be configured to operate as one or more software modules to perform the operations of one embodiment of the present invention, and vice versa.

In the present invention as described above has been described by the specific embodiments, such as specific components and limited embodiments and drawings, but this is provided to help the overall understanding of the present invention, the present invention is not limited to the above embodiments, Various modifications and variations can be made by those skilled in the art to which the present invention pertains. Therefore, the spirit of the present invention should not be limited to the described embodiments, and all the things that are equivalent to or equivalent to the claims as well as the following claims will belong to the scope of the present invention. .

Claims

In the apparatus for predicting the pleasure-discomfort index of a word,

Calculating an emotional correlation of each of the word and one or more comparison words, calculating an emotional correlation of each of the at least one reference word and each of the at least one reference word preset, and an emotional similarity of each of the word and the at least one reference word A calculating unit calculating a; And

A predicting unit predicting the pleasant-discomfort index of the word by using the at least one emotional similarity and the preset pleasant-discomfort index of the at least one reference word;

Pleasantness-discomfort index prediction device of the word comprising a.
The method of claim 1,

The calculation unit,

The word or the probability using the probability of the word or the reference word and the comparison word appearing independently in a paragraph, and the probability of the word or the reference word and the comparison word appearing together in a paragraph; Calculating an emotional correlation between the reference word and the comparison word.
The method of claim 1,

The calculation unit,

Calculating a first vector having an element associating emotions of each of the word and the at least one comparison word, and a second vector having an element associating each of the at least one reference word and each of the at least one comparison word; And an emotional similarity of each of the word and the at least one reference word using an angle between the first vector and the second vector.
The method of claim 1,

The predicting unit predicts the pleasant-discomfort index of the word using the following equation.

Where P / UNPp is the pleasant-to-off index of the predicted word, SUM is the sum of the emotional similarities of the word and the at least one reference word, n is the index of the at least one reference word, and N is the at least one The number of reference words, FSn is the emotional similarity between the word and the nth reference word, and P / UNPn means a preset comfort-discomfort index of the nth reference word, respectively.
In the method of predicting the pleasure-discomfort index of a word,

Calculating an emotional association of each of the words and one or more comparison words;

Calculating an emotional correlation of each of at least one preset reference word and each of the one or more comparison words;

Calculating an emotional similarity of each of the word and the at least one reference word; And

Predicting the pleasure-discomfort index of the word using the at least one emotional similarity and a preset pleasure-discomfort index of the at least one reference word;

Pleasant-discomfort index prediction method of a word comprising a.
A computer-readable recording medium having recorded thereon a program for performing the method of claim 5.