Keywords

1 Introduction

Sentimental analysis, emotional and affective computing have gained special attention during the last years as the improvements that have been achieved in the recognition, interpretation, processing, and simulation of human affects, which are required abilities for many HCI applications [2, 3]. Particularly, sentimental analysis also known as opinion mining employs methods from natural language processing, text analysis and computational linguistics to interpret information from a given source. It has successfully been used on social networks to extract useful information specially for instance for customer services and marketing concerns.

In this paper, we present an hybrid sentimental-content analysis in order to improve the recognition and interpretation of human affects. Content analysis is a research technique that has extensively and fruitfully been used in the computational linguistic field for the objective, systematic and quantitative description of a given communication. We employ as case study the #NiUnaMenos (Not One Less) social movement, which demands for an end to femicide and violence against women. We select this case as several human affects are involved. This movement started on Argentine and was rapidly viralyzed to Latin America and worldwide as a powerful sign of protest in order to stop misogyny. We gather a corpus composed of tweets using the #NiUnaMenos hashtag. This corpus is firstly analysed by state-of-the-art sentiment analysis algorithms, employing 10 different levels of human affects from fully negative to fully positive in order to improve the certainty of results. This output is then re-evaluated by content analysis so as to refine it and eliminate inconsistencies. Interesting results are obtained and discussed about the use of this hybrid in order to the correct recognition and interpretation of human affects specially for the emotional strength of the analyzed study case and in general for any communication context.

The remainder of this paper is organized as follows: Next section provides the analysis and results followed by conclusions and some directions of future work.

2 Analysis and Results

We have gathered a corpus of about 300 tweets that used the #NiUnaMenos hashtag from randomly selected accounts in order to avoid an unfair evaluation. We firstly employ the SentiStrength [4] tool to pre-process the data. SentiStrength performs automatic sentimental analysis on texts by estimating the strength of positive and negative sentiment in parallel as human being does [1]. It employs a range from −1 (not negative) to −5 (extremely negative) and from 1 (not positive) to 5 (extremely positive). Table 1 details the values encountered by SentiStrength for the analyzed tweets. After this pre-processing phase we analyze the data via content analysis.

Table 1. Strength of positive and negative sentiment
Full size table

Before starting the content analysis, we may observe that most of tweets fall into the “not negative” and “not positive” category. Moreover, the tool did not tagged any tweet with −5, situation which seems to be very positive. We may think that only using the information given by the pre-processing we can have a general opinion of the whole context. However, applying only sentimental analysis is not completely feasible this purpose and it is necessary to deeply analyze the content by a second technique.

When applying content analysis, we observed several tweets that caught our attention, we highlight some of them since our opinion is that its content is extremely negative, however no −5 value is encountered by the sentimental analysis pre-process.

  • RT @elliberalweb: #NiUnaMenos. La asesinaron a golpes delante de su hija y quemaron la casa. (She was beaten to death in front of her daughter and her house was burned).

In our opinion this tweet should be tagged as “extremely negative” due to its hard content including words with clear negative strength such as “beaten”, “burned” and “death”. A similar situation occurs with the following two tweets.

  • RT @ARGNoticiasok: Se arrojó de un taxi en movimiento para evitar abuso sexual #Inseguridad #Taxi #NiUnaMenos ... (He jumped from a moving taxi to avoid sexual abuse #Insecurity #Taxi #NiUnaMenos...).

  • RT @LaAlamedaMor: #NiUnaMenos asesinada, violada, desaparecida, levantada, prostituida, golpeada, discriminada #Feminicidios #Morelos (#NiUnaMenos murdered, raped, disappeared, raised, prostituted, beaten, discriminated #Feminicides #Morelos).

Then, from another standpoint, the following tweet as well as previous ones have negative strength, however we may think that those tweets are positive since they support the #NiUnaMenos movement.

  • RT @vervemediaes: No más pérdidas de identidad, no más miedo, humillaciones, insultos y palizas. #NiUnaMenos hay salida (No more loss of identity, no more fear, humiliation, insults and beatings. #NiUnaMenos there is an exit).

Finally, we can also detect ironies on some tweets, which are clearly teasing movement such as the tweets stated below. This is hard to detect for a sentimental analysis tool.

  • RT @FeliLoGlobo: Me comera 8 medialunas, #NiUnaMenos (I would eat 8 croissants, #NiUnaMenos).

  • RT @ElGallo_ar: @dzapatillas @inadi Las #NiUnaMenos jamas mencionan estas desigualdades, que se jubilen a los 65 como los hombres (The #NiUnaMenos never mention these inequalities, they must retire at 65 like men).

We have applied the complete content analysis by hand to most negative and positive tweets, for space reasons we do not include the whole analysis in this paper, but we can conclude that while the sentiment analysis gives us valuable data, it is still necessary to apply a manual technique in order to have access to the richness of those messages and correctly interpret the context where they are placed.

3 Conclusions and Future Work

In this paper we have studied the combination of sentimental and content analysis, sentimental analysis allows one to systematically study affective states by using natural processing techniques, while content analysis focuses on systematic and quantitative description of a given communication form. We have employed as case study the #NiUnaMenos (Not One Less) social movement due to its emotional strength. We have collected a set of tweets, which have firstly been pre-processed by using the SentiStrength sentimental analysis tool and then studied by content analysis. We may conclude that the automatic sentimental analysis is a powerful tool for alleviating the manually content analysis process. However, it is hard to have a full evaluation by only using this pre-process. Richer information is gathered when after the pre-processing, another content analysis technique is employed. As future work we aim at deeply analyzing this combination by exploring other similar thematics strongly involving human affects.