KR20190023721A - Apparatus and method for gathering public mood about place - Google Patents

Abstract

An apparatus for inferring and standardizing public sensitivity for a detailed place comprises: a placeness extraction unit extracting effective data including information related to a detailed place becoming a purpose of online social data including information related to a place; and a detailed place atmosphere standardizing module extracting public sensitivity of the corresponding detailed place of the effective data.

Description

[0001] APPARATUS AND METHOD FOR GATHERING PUBLIC MOOD ABOUT PLACE [0002]

The present invention relates to a technique for analyzing user opinions posted on online social media.

Opinion generally consists of an opinion target, an expressed positive / negative sentiment, and an opinion holder [Liu 12]. The nominal object of opinion exists as a specified entity and contains a set of various attribute / features whose subject of expression is to be commented [Liu06]. This way of analyzing opinions on any object is called pinion mining, or entiment analysis.

Opinion mining techniques are largely classified into a bag-of-words method (LP-based approach), which considers the subject to be analyzed as a set of words without considering the order and structural characteristics . It can be divided into sentence-level, document-level, and aspect-level opinion mining depending on what is analyzed. Sentence unit analysis analyzes what kind of subject, what object, and what sensitivity from each sentence. The document unit analysis attempts to understand how one or more subject expresses a certain emotion from a document composed of sentences, and the attribute unit Opinion Mining performs an emotion analysis on the properties of each object.

Early opinion mining technology was mainly focused on the extraction of one representative sentiment from the viewpoint of document level [Pennebaker01]. However, since the texts expressed by people through online social media and the web are expressed in short form and in various forms, they have developed as sentence-level research to understand this [Esuli06]. Furthermore, attempts have been made to analyze the emotional properties of the attribute units to understand the opinions expressed by people on certain characteristics of a particular object, thereby enabling opinion mining on various features of one object [Blair08].

In order to extract such subdivided opinions, a set of keywords that describe the characteristic aspect of the object is needed. However, when opinion mining is performed using words defined on general topics, accuracy is high but its coverage is very low [Glasgow14]. Thus, various methods have been attempted to construct a set of keywords with the vocabulary that people primarily use for a particular subject. In the beginning, a method of adding dictionary-based keywords based on the dictionary meaning of words has been attempted, but this method has limitations that it is difficult to effectively extract emotions because the coverage can be increased but people can not include words actually used . Therefore, methods for expanding commonly used keywords starting from a generalized set of a few keywords have been devised. Typical examples are Pointwise Mutual Information (PMI) based [Bouma09] and Graph-based [Blair08] methodologies. . Recently, a method for expanding a set of keywords using crowdsourcing has been attempted, but this has the disadvantage that it requires additional cost and it is difficult to guarantee the reliability of the result [Hutto14].

An object of the present invention is to provide an apparatus and method for stereotyping emotional characteristics of on-line information related to a detailed place and extracting a formalized detailed place atmosphere.

The apparatus of the present invention includes a data collecting unit, a morpheme analyzing unit, a word embedding module, a clustering module, a locality extracting unit, and a detailed site atmosphere shaping module.

The data collection unit collects various types of data that are disclosed on-line, such as social media, blogs, and news, which contain information related to the location of the detailed place, and stores the collected data in the form of contents and metadata.

The Morpheme Analysis Department morphs and formalizes location-based social data.

The word embedding module comprehends emotional vocabularies in the place-based social data in an automated manner to formulate the public emotional vocabularies related to the place-locality, receives the morpheme-analyzed place-based social data, and outputs the vector-shaped words .

The clustering module understands the relationships between the words in place social social data output from the word embedding module.

The location property extraction unit extracts the vocabulary representing the formalized place property, extracts the action vocabulary related to the location based on the formalized locality system, and uses the location information based on the GPS or the noun information received from the morphological analysis unit, And the location information of the detailed place is extracted by associating proper nouns (entities).

The detailed place atmosphere stereotyping module formalizes the atmosphere of the detailed place corresponding to the place property by using the clustered words obtained from the clustering module and the predefined place atmosphere related words. The detailed place atmosphere stereotyping module shapes the words belonging to the same community, which are located close to the corresponding word, with the atmosphere word related to the place, using the atmosphere word as the seed word.

According to the embodiment, an intelligent personal secretary service optimized for various requirements of users is solved by solving the problem that only the physical location that lacks various meaning consideration of the place, which is the limitation point of current intelligent personal secretary service, Can be provided. According to the embodiment, it is possible to enhance user satisfaction by enhancing real-time and personalization in the field of intelligent personal assistant services having various forms and scalability.

According to an embodiment, an intelligent personal assistant, a mobile service, and a location-based information / content recommendation service industry can provide a detailed place atmosphere.

According to the embodiment, the social commerce market can also be entered through a process of recommending suitable detailed popularity place and contents, and issuing customized advertisement and mobile coupon.

FIG. 1 is a block diagram of an emotional characteristic shaping apparatus of on-line information related to a detailed place.

The regional emotional tendency extraction study is as follows.

[Bertrand13, Mitchell13] The study used geo-tagged posts (including geo-tagged) among social media platform Twitter data to identify the regional emotions of the region. Online postings, which mainly include location information produced by smart devices, can achieve high resolution emotional maps because they can get accurate coordinates through GPS. In addition, since the update rate of data is high and it includes the accurate generation time information, it is possible to analyze the emotional trend by the day of the week or the time, and has a possibility to be applied to real-time emotion extraction in the future. In the [Brereton08] study, regional differences were analyzed from a more detailed viewpoint compared to the simplified emotion information of positive / negative such as happiness. The [Rentfrow13] study attempted to extract the emotional mood of each region of the United States based on the Big five personality types (openness, integrity, extroversion, affinity, neuroticism). This method has the advantage of trying to measure subtle emotional characteristics compared to the previous studies that have only tried to extract the emotional sensation by region.

Although there have been studies to extract regional emotions as described above, there are the following limitations. First, the characteristics of opinion holders with emotions are not considered. Collective emotional analysis that does not take into account the characteristics of the subjects has been performed. The present invention proposes emotional analysis that can consider individual characteristics because it is a mining technique for real-time intelligent information recommendation service in smart space .

Second, the above studies analyzed emotions using only the most general keyword sets that express emotions, not the set of keywords used by people expressing actual emotions. A common set of emotional keywords can be high in accuracy, but its coverage may not be enough to understand the emotions exposed in the actual text.

Third, it is inappropriate to be the subject of emotional analysis for the recommendation system because the area to be analyzed is a high level place such as county or state.

Finally, the above-mentioned studies analyzing sensibility using Opinion Mining technology have a limitation that analysis is performed only on simple dimensions such as positive / negative. Therefore, studies such as [Brereton08, Rentfrow13] attempt to understand the emotions of various dimensions, but this is based on the questionnaire and is inadequate to be used in recommendation systems.

Accordingly, the present invention provides an apparatus and method for shaping emotional characteristics of on-line information related to detailed places in order to solve these limitations.

The apparatus and method of the present invention can quantitatively understand public sentiment from various expressions used by people, and public sentiment in place social media can be extracted for various dimensions beyond positive / negative.

For the place to be the object of the recommendation, understanding of the atmosphere from the public sensitivity and it should be done. In order to be used in the recommendation system, emotional analysis should be performed on appropriate places having place characteristics. For this purpose, understanding of the atmosphere from the public emotion should be preceded.

The emotional characteristics of the place should be considered in order to recommend a place suitable for the user's detailed purpose.

In the existing recommendation algorithm, the emotional element was used because it was difficult to apply the emotional context to the individual user because of lack of understanding of the context and the subject of expression. When the present invention extracts emotions about the detailed situation and context of the place, it is necessary to utilize the emotional characteristics of the place in relation to the detailed purpose of the user. In addition to the personal preference and taste information of the user, the emotional characteristic of the place represented in the user and the situation most suitable to the condition is analyzed considering the detailed purpose of visiting the place, and based on this, I can reason. That is, if the user's detailed purpose is determined, the emotional characteristic expressed by the user similar to the user in relation to the target can be extracted and used for the recommendation service based on the extracted emotional characteristic.

The present invention will be described with reference to Fig.

FIG. 1 is a block diagram of an emotional characteristic shaping apparatus of on-line information related to a detailed place.

Since online social data exist in the form of unstructured data created by people, there is a limit to applying existing emotional analysis techniques. Therefore, the online social data specialized emotional analysis related to the detailed place is supplemented as shown in Fig.

The apparatus includes a data collecting unit, a morpheme analyzing unit, a word embedding module, a clustering module, a locality extracting unit, and a detailed site atmosphere shaping module, and extracts emotion information from the place-based social data and formats it.

The data collection unit collects various types of data that are disclosed on-line, such as social media, blogs, and news, which contain information related to the location of the detailed place, and stores the collected data in the form of contents and metadata.

The Morpheme Analysis Department morphs and formalizes location-based social data. The morpheme analysis unit receives the place-sex social data as input and obtains the sequence of the morpheme-analyzed word. Morpheme analysis can take advantage of various open sources that have already been implemented and can utilize different analytical sources according to location-based social data. For example, to analyze Hangul, we use a morpheme analyzer, Hananum analyzer, and Stanford NLP tool for English analysis.

The word embedding module understands the emotional vocabularies in the place-based social data in an automated way in order to formalize the popular emotional vocabularies related to the place nature. To do this, we embed the word embedding. Word embedding is learning the relationship of words in a document in Distributed Semantics. To implement this, we can use word2vec algorithm to learn the relation between words in vector form. The Word2vec algorithm was created by Google's Tomas Milkov. It uses the Skip grams to predict the probability of the surrounding words (context) c appearing when the word w is given, and when the context c is given using CBOW Predicts the probability of occurrence of the word w. The module receives the morpheme-analyzed location-based social data as input and outputs the words in vector form.

The clustering module uses the word embedding module to understand the relationships between words in place-based social data. Because words are expressed in vector form, we use vector distance calculations to understand which words have similar meanings, or which words are not related at all. Furthermore, it is possible to understand the grouping of words in some form by using a distance-based clustering algorithm. For implementation, hard clustering clustering such as k-means or soft clustering method such as Mixture of Gaussian is applied and Euclidean distance can be used for vector distance calculation. The module will result in a cluster of closely related words.

The locality extraction unit extracts the place locality from the locality social data to obtain the emotional vocabulary related to locality. To this end, the locality extraction unit extracts a vocabulary representing stereotyped locality. The locality extractor extracts the action vocabulary related to the locality using a formalized locality system. Receives location-based social data that is morpheme-typed as input, and outputs place-based words for given place-based social data as output. If the collected data includes meta data (geotag) that directly contains the location information, the GPS based location information is used together, and if the location information is missing, it is possible to extract the location information of the detailed place by associating the entity.

The detailed place atmosphere stereotyping module formalizes the atmosphere of the detailed place which is finally the place property by using the clustered words obtained from the clustering module and the words related to the appointed place atmosphere. The atmosphere word is used as a seed word, and words belonging to the same community located at a distance from the corresponding word are stereotyped with atmosphere characteristics related to the place.

In this way, the emotional characteristics of the place are deduced from the place-related text data collected through on-line social media and utilized to support the customized recommendation service. Emotional analysis techniques have been used to collect subjective opinions of a certain object, but emotional elements in existing recommendation algorithms are not applied to individual users because of lack of understanding of subject and context in which emotion is expressed. Was extremely difficult to use. In addition, for user-customized recommendation service, the emotional characteristics should be evaluated in various characteristic aspects of the analysis target site. In order to extract the refined results, it is necessary to extract the atmosphere specific to the detailed place to analyze. Therefore, the present invention considers the multi-class mood of the place and the place of the object to be expressed in addition to the opinion target and the sentiment of the emotion, which are elements of the conventional emotional analysis technique, Emotional analysis for recommendation service support can be done.

Claims (4)

As an apparatus for inferring and shaping public sensitivity to a detailed place,
A location property extraction unit that extracts, from among online social data including information related to a place, valid data including information related to a target detailed location, and
A detailed location atmosphere shaping module for extracting the public sensibility of the detailed site from among the valid data
Wherein the emotional reasoning and shaping device comprises: The method of claim 1,
Wherein the locality extraction unit comprises:
Wherein the detailed location is extracted using GPS-based location information or location-related keyword information. The method of claim 1,
Wherein the detailed location atmosphere shaping module comprises:
Emotional reasoning and shaping device that performs the morphological analysis, word embedding, and clustering process on the collected data as preprocessing. The method of claim 1,
Wherein the detailed location atmosphere shaping module comprises:
A sensory inference and stereotyping device for searching representative clusters of the closest word based on the detailed place extracted by the locality extraction unit, and formalizing the representative sensibility of the public related to the detailed place using the predefined atmosphere related wordbook.

Images