Search

PurposeThis study aims to compare online review characteristics, review length and review sentiment score between “organic” and “regular” food products. In addition, variations in the consumer sentiment scores across the review lengths are studied.Design/methodology/approachThis study fits into the descriptive research design. From Amazon’s website, the consumer product reviews are scrapped. Using the text analytical package “sentiment” in R-Studio, we computed the sentiment scores and counted the number of words in each review. The mean sentiment scores and mean review length are compared for regular and organic products using one-way ANOVA. Sentiment score variation across review length and product class is studied through factorial ANOVA. Sample reviews of ghee and honey are used to test the hypotheses.FindingsThe review length shows a significant difference between the regular and organic products. The mean number of words in the regular products reviews is significantly lower than the mean number of words in the organic product reviews. The regular products’ mean sentiment score is significantly lower than the mean sentiment score of organic products. The mean sentiment scores are not consistent between ghee and honey. Sentiment scores are better for organic honey and regular ghee products. For regular ghee products, longer reviews result in lower sentiment scores. For regular and organic versions of honey, longer reviews are associated with better sentiment scores.Research limitations/implicationsThis study did not include the helpfulness of a review and the demographic data of the reviewers.Practical implicationsSentiment scores’ variations across the regular and organic and product categories such as ghee and honey give a comprehensive feedback to the firms. It also indirectly communicates a brand’s evaluation by the consumers and the performance feedback for an upward extension like the organic category.Social implicationsStudies on organic category give feedback for environment-friendly products and consumer attitude shift towards safer products.Originality/valueVery limited studies have reported the upward line extensions. The authors study the upward line extension organic and associated sentiment scores variation. The role of review length and its systematic influence on the sentiment scores, variations in the review due to the product nature (organic/regular) are unique contributions of this study.

Understanding consumer’s risk perception on the Belt and Road (B&R) countries is important for the development of cross-border e-commerce (CBEC) along these countries. However, most of the extant studies cannot properly analyze consumer’s risk perception due to the limited data collected by questionnaires. Therefore, this study proposes a text-mining-based framework to study consumer’s risk perception on the B&R countries based on massive textual online reviews collected from CBEC. In the proposed framework, the Latent Dirichlet Allocation model and sentiment analysis method are used to identify the main risk factors affecting consumer’s risk perception and calculate their sentiment score, the risk perception indicator is constructed to measure the magnitude of consumer’s risk perception. In the experiment, totally 66,661 reviews of the representative products from nine B&R countries are collected from Tmall Global. Six major risk factors are identified, and consumer’s risk perception on nine B&R countries is given.

Gamification is here to stay, and tourism and hospitality online review platforms are taking advantage of it to attract travelers and motivate them to contribute to their websites. Yet, literature in tourism is scarce in studying how effectively is users’ behavior changing through gamification features. This research aims at filling such gap through a data-driven approach based on a large volume of online reviews (a total of 67,685) collected from TripAdvisor between 2016 and 2017. Four artificial neural networks were trained to model title and review's word length, and title and review's sentiment score, using as input 12 gamification features used in TripAdvisor including points and badges. After validating the accuracy of the model for extracting knowledge, the data-based sensitivity analysis was applied to understand how each of the 12 features contributed to explaining review length and its sentiment score. Three badge features were considered the most relevant ones, including the total number of badges, the passport badges, and the explorer badges, providing evidence of a relation between gamification features and traveler's behavior when writing reviews. info:eu-repo/semantics/acceptedVersion

Aiming at social media texts with weak grammatical features, based on the weak-rule text and word vectors, this paper proposes a framework combining sentiment lexicon and neural network. Moreover, we study the influence of sentiment lexicon on improving the accuracy of text sentiment analysis. According to the structure of the word vector and the uniqueness of the sentiment score in the sentiment lexicon, our model first converts the text into a sentiment score vector. Secondly, the converted sentiment score vector is used to learn its hidden emotional representation through the neural network. At the same time, we use a combination of pre-training word vectors and the BiLSTM network to learn the hidden semantic representation of the original text. Finally, the hidden emotional representation and the hidden semantic representation are combined to analyze the weak-rule text sentiment. Experimental results on Weibo text with weak grammatical rules show that the performance of the method which use the sentiment score vector is better than traditional sentiment lexicon methods. The performance of combining hidden emotion representation and hidden semantic representation is also better than the BiLSTM that uses pre-trained word vectors and attention mechanism.

Online business is one of the rapidly growing business sectors of current world. Now-a-days people purchase a lot of things from online shopping sites. Sales of online products are most often review driven. Thus, detecting deceptive reviews is getting more importance day by day. Sentiment analysis has great importance in fake review detection system. This paper introduces a sentiment analysis model that can separate positive and negative sentimental reviews efficiently. It shows an analysis of sentiment distribution for fake and truthful reviews. Also, the proposed sentiment model is used to find the impact of probabilistic sentiment score in fake online review detection using a hotel review dataset.

Structured data on customer feedback is becoming more costly and timely to collect and organize. On the other hand, unstructured opinionated data, e.g., in the form of free-text comments, is proliferating and available on public websites, such as social media websites, blogs, forums, and websites that provide recommendations. This research proposes a novel method to develop a knowledge-based recommender system from unstructured (text) data. The method is based on applying an opinion mining algorithm, extracting aspect-based sentiment score per text item, and transforming text into a structured form. An action rule mining algorithm is applied to the data table constructed from sentiment mining. The proposed application of the method is the problem of improving customer satisfaction ratings. The results obtained from the dataset of customer comments related to the repair services were evaluated with accuracy and coverage. Further, the results were incorporated into the framework of a web-based user-friendly recommender system to advise the business on how to maximally increase their profits by introducing minimal sets of changes in their service. Experiments and evaluation results from comparing the structured data-based version of the system CLIRS (Customer Loyalty Improvement Recommender System) with the unstructured data-based version of the system (CLIRS2) are provided.

The purpose of social media websites like Twitter, Tumbler, and Facebook is that its user can express their feelings without being pressurized by anyone. User can give their point of view regarding the recent events in their surroundings as well as give suggestions to improve surroundings in text-based format while conveying their emotions which they are not able to easily verbalize using emoticons and emoji. For better understanding of people’s opinion, it is important to analyze this semiotics as well as sentence. In this paper we will discuss importance of semiotics in sentiment analysis. The main contribution of this paper to provide an approach to determine sentiment score of a tweet with semiotics with multi-dimensional sentiment analysis. In our algorithmic approach we have created semiotic dictionary which have sentiment score for each semiotic with sentiment expressed by it most frequently. We have compared our algorithmic approach with the prediction approach for sentiment classification and calculating sentiment scores. Proposed approach overcome limitation of regression analysis approach as it also helps finding sentiment score in case of where semiotic role is “Addition” and it is more effective at calculating sentiment score than other approach.

Sentiment analysis is an extension of data mining which employs natural language processing and information extraction task to recognize people's opinion towards entities such as products, services, issues, organizations, individuals, events, topics, and their attributes. It gives the summarized opinion of a writer or speaker. It has received lot of attention due to increasing number of posts/tweets on social sites. The proposed system is meant to classify a given text of review into positive, negative, or the neutral category. Primary objective of this article is to provide a method of exploiting permutation and combination and chi values for sentiment analysis of product reviews. Publicly available freely dictionary SentiWordNet 3.0 has been used for review classification. The proposed system is domain independent and context aware. Another objective of the proposed system is to identify the feature specific intensity with which reviewer has expressed his opinion. Effectiveness of the proposed system has been verified through performance matrix and compared with other research work.

Recommender system is a strategy in e-commerce, which recommends items based on the user’s interest. It has the capability to predict whether a particular user would prefer an item or not based on the user’s profile. Recommender systems are useful for both e-commerce service provider and users. So it should be required for a recommendation system to provide most preferable items to the user’s interest. This paper presents a dynamic recommendation system to provide recommendations on the user’s interest. In this dynamic recommendation system first of all the web usage information is utilized to find the user’s behavior and then similar user behavior score is computed. In the second phase, product information is collected on the basis of current search information about the user through which sentiment score and social media popularity score are computed. On the other side coefficient matrix is used to calculate user’s purchasing power. These three factors- similar score behavior, sentiment score and popularity score are used to calculate the combined weight for the particular product. Then a coefficient matrix and the computed weights are used to calculate the possible recommendation of the product for the user.

Purpose The purpose of this study is to measure the interaction between media sentiment and the Bitcoin price. Because some researchers argued that the Bitcoin value is also determined by perception of users and investors, this paper examines how. Design/methodology/approach The database of relative news articles as well as blog posts has been collected for the purpose of this research. Hence, each article has been given a sentiment score depending on the negative and positive words used in the article. Findings This paper has identified that interaction between media sentiment and the Bitcoin price exists, and that there is a tendency for investors to overreact on news in a short period of time. Originality/value While sentiment analysis of Twitter posts as a predictor of the Bitcoin price has been conducted in the past, this research does not have any analog because psycho-semantic dictionaries have not been applied earlier in the Bitcoin research.

Twitter behaves as a social sensor of the world. The tweets provided by the Twitter Firehose reveal the properties of big data (i.e. volume, variety, and velocity). With millions of users on Twitter, the Twitter's virtual communities are now replicating the real-world communities. Consequently, the discussions of real world events are also very often on Twitter. This work has performed the real-time analysis of the tweets related to a targeted event (e.g. election) to identify those potential sub-events that occurred in the real world, discussed over Twitter and cause the significant change in the aggregated sentiment score of the targeted event with time. Such type of analysis can enrich the real-time decision-making ability of the event bearer. The proposed approach utilizes a three-step process: (1) Real-time sentiment analysis of tweets (2) Application of Bayesian Change Points Detection to determine the sentiment change points (3) Major sub-events detection that have influenced the sentiment of targeted event. This work has experimented on Twitter data of Delhi Election 2015.

The payday lending industry has grown dramatically over the past two decades, with lenders increasingly marketing their products on the Internet. However, little is known about payday loan (PDL) marketing through social media networks. Our study is the first to report the PDL promotions on Twitter. Out of 23,276 tweets related to PDLs from June 29, 2015 to October 2, 2015, 71% were commercial tweets, most of which came from a small percentage of users. The overall impact to the general public was large and we observed a range of opinions in our sentiment analysis. Commercial tweets had an average sentiment score of 34.9 while noncommercial had a total average sentiment score of 50.7. Commercial tweets were more concentrated in states with restrictive or hybrid regulations on PDL. Our findings provide evidence for enhancing online PDL regulations and using social media data for real-time monitoring of PDL lending.

As violent extremists continue to surface in online discussion forums, law enforcement agencies search for new ways of uncovering their digital indicators. Researchers have both described and hypothesized a number of ways to detect online traces of potential extremists, yet this area of inquiry remains in its infancy. This study proposes a new search method that, through the analysis of sentiment, identifies the most radical users within online forums. Although this method is applicable to web-forums of any type, the method was evaluated on four Islamic forums containing approximately 1 million posts of its 26,000 unique users. Several characteristics of each user’s postings were examined, including their posting behavior and the content of their posts. The content was analyzed using Parts-Of-Speech tagging, sentiment analysis, and a novel algorithm called ‘Sentiment-based Identification of Radical Authors’, which accounts for a user’s percentile score for average sentiment score, volume of negati...

In present research, the authors examined how social media influencers affect the overall sentiment of a topic. To this end, they utilized supervised machine learning approach to develop SentiRobo for measuring the sentiment score of social media content. In the next stage, they studied social media datasets with 375,141 records in the education domain to investigate the correlation between social media topics and top authors' sentiment. The Pearson correlation test results revealed that top one percent of social media authors are enough to significantly influence the whole sentiment of each topic.

The using of Twitter by selebrities has become a new trend of impression management strategy . Mining public reaction in social media is a good strategy to obtain feedbacks, but extracting it are not trivial matter. Reads hundred of tweets while determine their sentiment polarity are time consuming . Extractive sentiment summarization machine are needed to address this issue. Previous research generally do not include sentiment information contained in a tweet as weight factor, as a results only general topics of discussion are extracted. This research aimed to do an extractive sentiment summarization on both positive and negative sentiment mentioning Indonesian selebrity, Agnes Monica , by combining SentiStrength, Hybrid TF-IDF, and Cosine Similarity. SentiStrength is used to obtain sentiment strength score and classify tweet as a positive, negative or neutral. The summarization of posisitve and negative sentiment can be done by rank tweets using Hybrid TF-IDF summarization and sentiment strength score as additional weight then removing similar tweet by using Cosine Similarity. The test results showed that the combination of SentiStrength, Hybrid TF-IDF, and Cosine Similarity perform better than using Hybrid TF-IDF only, given an average 60 % accuracy and 62% f-measure . This is due to the addition of sentiment score as a weight factor in sentiment summ­ari­zation.

Previous studies carried out customer surveys by questionnaires to collect data for analyzing consumer requirements. In recent years, a large and growing body of literature has investigated the extraction of customer requirements and preferences from online reviews. However, since customer requirements change dynamically over time, traditional studies failed to obtain the change data of customer requirements and opinions based on sentiments expressed in reviews. In this paper, a new method for dynamically mining user requirements is proposed, which is used to analyze the changing behavior of product attributes and improve product design. Dynamic mining differs from the traditional need acquisition mainly in three aspects: (1) it involves dynamically mining user requirements over time (2) it adds changes in manufacturers’ opinions to the analysis (3) it allows for product improvement strategies based on the changing behavior of product attributes. First, text mining is adopted to collect customer and manufacturer review data for different time periods and extract product attributes. A Natural Language Processing tool is used to measure the importance weight and sentiment score of product attributes. Second, an approach for dynamically mining user requirements is introduced to classify product attributes and analyze the changes of attribute data in three categories over time. Finally, an improvement strategy for next-generation product design is developed based on the changing behavior of attributes. Moreover, a case study on vehicles based on online reviews was conducted to illustrate the proposed methodology. Our research suggests that the proposed approach can accurately mine customer requirements and lead to successful product improvement strategies for next-generation products.

The web provides a suitable media for users to share opinions on various topics, including consumer products, events or news. In most of such content, authors express different opinions on different features (i.e., aspects) of the topic. It is a common practice to express a positive opinion on one aspect and a negative opinion on another aspect within the same post. Conventional sentiment analysis methods do not capture such details, rather an overall sentiment score is generated. In aspect based sentiment analysis, the opinions expressed for each aspect are extracted separately. To this aim, basically a two-phased approach is used. The first phase is aspect extraction, which is the detection of words that correspond to aspects of the topic. Once aspects are available, the next phase is to match aspects with the sentiment words in the text. In this work, we present a framework for the aspect based sentiment analysis problem on Turkish informal texts. We particularly emphasize the following contributions: for the first phase, improvements for aspect extraction as an unsupervised method, and for the second phase, enhancements for two cases, extracting implicit aspects and detecting sentiment words whose polarity depends on the aspect. Additionally, we present a tool including the implementations of the proposed algorithms, and a GUI to visualize the analysis results. The experiments are conducted on a collection of Turkish informal texts from an online products forum.

Online social networks have emerged as new platform that provide an arena for people to share their views and perspectives on different issues and subjects with their friends, family, relatives, etc. We can share our thoughts, mental state, moments, stand on specific social, national, international issues through text, photos, audio and video messages and posts. Indeed, despite the availability of other forms of communication, text is still one of the most common ways of communication in a social network. The target of the work described in this paper is to detect and analyze sentiment and emotion expressed by people from text in their twitter posts and use them for generating recommendations. We collected tweets and replies on few specific topics and created a dataset with text, user, emotion, sentiment information, etc. We used the dataset to detect sentiment and emotion from tweets and their replies and measured the influence scores of users based on various user-based and tweet-based parameters. Finally, we used the latter information to generate generalized and personalized recommendations for users based on their twitter activity. The method we used in this paper includes some interesting novelties such as, (i) including replies to tweets in the dataset and measurements, (ii) introducing agreement score, sentiment score and emotion score of replies in influence score calculation. (iii) generating general and personalized recommendation containing list of users who agreed on the same topic and expressed similar emotions and sentiments towards that particular topic. (C) 2019 Elsevier B.V. All rights reserved.

Occupant comfort and satisfaction in residential buildings is often subpar, yet traditional post-occupancy evaluation surveys are time-consuming and difficult to achieve large samples. Moreover, quantitative results do not necessarily provide insights about how to improve indoor environmental quality (IEQ). To address these limitations, this paper proposes a novel method to develop new knowledge about occupant comfort and satisfaction: text-mining of public guest reviews in temporary accommodations. Using a set of 1.35-million Canadian Airbnb reviews, a methodology is developed and demonstrated to obtain top reported causes for IEQ complaints, assess seasonable trends of IEQ issues, and quantify the frequency of multi-domain IEQ complaints. The results indicate that five percent of all reviews complained of IEQ issues, while a quarter of a percent complained of multiple forms of IEQ. Reviews with IEQ complaints had a statistically significantly worse overall sentiment score, suggesting significant importance of IEQ on overall guest satisfaction. Overall, the method yielded new quantitative and qualitative knowledge about IEQ in guest homes, but the developed text-mining methods have some limitations, such as failing to correctly interpret idioms and distinguishing comfort-related words that have multiple meanings.

Traditional Collaborative filtering algorithm works by using only the past experience of a user. To overcome the limitations of the traditional collaborative algorithm, an item based collaborative filtering system was introduced. In this paper, an improved recommender system is proposed. A dictionary of sentiment scores is created. These sentiment scores are calculated by finding the probability of the reviews to be positive. This sentiment score is used by an item based collaborative filtering system to improve the recommendations and filter out items with overall negative user opinion. The performance of the proposed system is compared with previous work done in this field.

Aspect-based financial sentiment analysis, which aims to classify the text instance into a pre-defined aspect class and predict the sentiment score for the mentioned target. In this paper, we propose a neural network model, Attention-based LSTM model with the Aspect information (ALA), to solve the financial opinion mining problem introduced by the WWW 2018 shared task. The proposed neural network model can adapt to the financial dataset so that the neural network can effectively understand the semantic information of the short text. We evaluate our model with the 10-fold cross-validation, and we compare our model with a variety of related deep neural network models.

Traditional methods of annotating the sentiment of a document are based on sentiment lexicons, which have been proven quite efficient. However, such methods ignore the effect of supplementary features (e.g., negation and intensity words), while only consider the counts of positive and negative words, the sum of strengths, or the maximum sentiment score over the whole document primarily. In this paper, we propose to use convolutional neural network (CNN) and long short-term memory network (LSTM) to model the role of negation and intensity words, so as to address the limitations of lexicon-based methods. Results show that our model can not only successfully capture the effect of negation and intensity words, but also achieve significant improvements over state-of-the-art deep neural network baselines without supplementary features.

Financial markets are usually affected by the news media which would allow us to consider the news sentiment effect on the market. In this study we propose the sentiment shock and sentiment trend. We provide an entropy based analysis for the two and compare them with the original sentiment score where we quantify the effectiveness of each sentiment in different financial sectors. We show that the sentiment shock and the sentiment trend are more efficient for cyclical sectors and both have a better capability to reduce the noise in the original signal around it's mean. Also, we design a dynamic sentiment shock and trend trading strategy with weekly evaluations where we show that a strategy based on sentiment shock or trend can have a superior performance compared to a buy-and-hold strategy in terms of risk-adjusted return.

With rapid development of E-commerce, nowadays more users preferred to purchase the product from on-line. To increase consumer satisfaction and on-line shopping practice, usually online merchants request their customers to reveal our opinions about the products which they had purchased. The reviews will be helpful for the customers to purchase a new product and it also helps the manufacturers to enhance the quality of the product. Aspect ranking of products aims in identifying and prioritizing the aspects of products based on the online reviews given by the customers. The aspects which are commented by the customers and the opinions on those aspects are considered to be the measures for ranking. We first extract the product aspects and then the opinions on those aspects are identified via sentiment dictionaries, finally the aspects are ranked based on their sentiment score.

News analysis activities have been the focus of many research studies across various life domains. So often, the goal of these studies is to automatically, analyze the meaning of news, and to gauge their impact on a particular domain. In this paper, we focus on studying sentiment analysis impact, on financial markets. Current studies, lack systematic approaches to evaluate the impact of a given sentiment dataset, in different financial contexts. We introduce a framework that encompasses models, processes, and a supporting software architecture for defining different financial contexts and conducting sentiment data-sets evaluation. The paper, describes a prototype implementation of the framework and a case study, which investigates the efficacy of the framework in evaluating the impact of a particular news sentiment dataset. The results demonstrate the capability of the framework in bridging the gap between producing a sentiment dataset and evaluating its impact in various financial contexts.

This paper describes a supervised solution for detecting the polarity scores of tweets or headline news in the financial domain, submitted to the SemEval 2017 Fine-Grained Sentiment Analysis on Financial Microblogs and News Task. The premise is that it is possible to understand market reaction over a company stock by measuring the positive/negative sentiment contained in the financial tweets and news headlines, where polarity is measured in a continuous scale ranging from -1.0 (very bearish) to 1.0 (very bullish). Our system receives as input the textual content of tweets or news headlines, together with their ids, stock cashtag or name of target company, and the polarity score gold standard for the training dataset. Our solution retrieves features from these text instances using n-gram, hashtags, sentiment score calculated by a external APIs and others features to train a regression model capable to detect continuous score of these sentiments with precision.

This paper demonstrates a Web service called TweetCric to uncover cricket insights from Twitter with the aim of facilitating sports analysts and journalists. It essentially arranges crowdsourced Twitter data about a team in comprehensive visualizations by incorporating domain-specific approaches to sentiment analysis.

We present a method that exploits syntactic dependencies for topic-oriented sentiment analysis in tweets. The proposed solution is based on supervised text classification and available polarity lexicons in order to identify the relevant dependencies in each sentence by detecting the correct attachment points for the polarity words. Our experiments are based on the data from the Semantic Evaluation Exercise 2015 (SemEval-2015), task 10, subtask C. The dependency parser that we used is adapted to this kind of text. Our classifier that combines topic- and sentence-level features obtained very good results.

Emotions are important in communication to effectively convey messages and to understand reactions to messages. Large scale studies of communication need methods to detect sentiment in order to investigate or model the processes involved. This chapter describes the sentiment strength detection program SentiStrength that was developed during the CyberEmotions project to detect the strength of sentiments expressed in social web texts. SentiStrength uses a lexical approach that exploits a list of sentiment-related terms and has rules to deal with standard linguistic and social web methods to express sentiment, such as emoticons, exaggerated punctuation and deliberate misspellings. This chapter also describes how SentiStrength can be refined for particular topics and contexts and how variants can be created for different languages. The chapter also briefly describes some studies that have applied SentiStrength to analyse trends in Twitter and You Tube comments.

Sentiment orientation data sets referred to variously as affective word lists, opinion lexicons, sentiment lexicons, emotion lexicons or sentiment dictionaries contain a list of words scored for the degree of positive and negative emotion they exhibit. Although these lists have been used extensively for the sentiment analysis of text data, they contain a limited number of words that are often inadequate for data obtained from modern text sources dominated by the influence of social media that has resulted in the creation and coining of new words on a regular basis. In an effort to enrich these data sets with new words, we propose two methods. The first method involves the sentiment analysis of portmanteau words. We have hypothesized that the sentiment score of a portmanteau word; which is a combination of two (or more) words and their meanings into a single new word; can be determined as a function of the sentiment scores of its component words. Regression analysis has been used to determine this functional relationship and several cases arising from the above have been evaluated on a data set constructed from SentiWordNet. The second method is an in situ approach for sentiment discovery for unknown words that uses labeled tweets and words from the sentiment orientation data set as inputs to discover the sentiment score of the unknown word. In order to validate the resultant score, we have also used a novel validation-feedback mechanism akin to cross-validation. Both these methods produce acceptable levels of accuracy proving that they can be implemented in practice.

Social networking applications are prominent among the internet user communities. Many social media websites are used for sharing the information instantly. Twitter is one of the vibrant social networking websites for sharing small textual information within a short span of time. It is essential to identify the type of information shared on these websites. Sentiment analysis involves the process of analyzing the opinion content present in the text. Millions of tweets are posted in a day about various topics. Twitter sentiment analysis mainly involves the process of identifying the polarity oriented terms mentioned in the tweet. Most of the twitter sentiment analysis works have concentrated on the sentiment polarity identification. Based on the literature, it is observed that, researchers still need to contribute in the area of sentiment score calculation of a tweet. Hence, in this work, sentiment score calculation is carried out with sentiment corpus oriented approach for calculating the score effectively. In addition, the grammatical type of the word used in a tweet, the relationship between the words are properly identified. The tweet tagger, corpus based sentiment score assignment have been distinctively used when compared to other previous works. The experimental results show that the sentimental score based tweet identification resulted in top tweets among the large collection of tweets.

The sentiment expressed by a meeting participant in their face-to-face comments may differ from the sentiment contained in their private summary of the meeting. In this work, we investigate whether we can predict the sentiment score of a participant's private post-meeting summary, based on multi-modal features derived from the group interaction during the meeting. We describe several effective prediction models, all of which outperform a baseline that assumes the sentiment score of the summary will be the same as the sentiment score of the participant's comments during the meeting.

Sentiment classification on clinical narratives has been a groundwork to analyze patient’s health status, medical condition and treatment. The work posed challenges due to the shortness, and implicit sentiment of the clinical text. The paper shows that a sentiment score of a sentence simultaneously depends on scores of its terms including words, phrases, sequences of non-adjacent words, thus we propose to use a linear combination which can incorporate the scores of the terms extracted by various language models with the corresponding coefficients for estimating the sentence’s score. Through utilizing the linear combination, we derive a novel vector representation of a sentence called language-model-based representation that is based on average scores of kinds of term in the sentence to help supervised classifiers work more effectively on the clinical narratives.

Gun related violence is a complex issue and accounts for a large proportion of violent incidents. In the research reported in this paper, we set out to investigate the pro-gun and anti-gun sentiments expressed on a social media platform, namely Twitter, in response to the 2012 Sandy Hook Elementary School shooting in Connecticut, USA. Machine learning techniques are applied to classify a data corpus of over 700,000 tweets. The sentiments are captured using a public sentiment score that considers the volume of tweets as well as population. A web-based interactive tool is developed to visualise the sentiments and is available at http://www.gunsontwitter.com. The key findings from this research are: (i) There are elevated rates of both pro-gun and anti-gun sentiments on the day of the shooting. Surprisingly, the pro-gun sentiment remains high for a number of days following the event but the anti-gun sentiment quickly falls to pre-event levels. (ii) There is a different public response from each state, with the highest pro-gun sentiment not coming from those with highest gun ownership levels but rather from California, Texas and New York., Comment: 10 pages, accepted to IEEE eScience 2016, Baltimore, USA

Sentiment analysis is an active field of research, moving from the traditional algorithms that operated on complete documents to fine-grained variants where aspects of the topic being discussed are extracted, as well as their associated sentiment. Recently, a move from traditional word-based approaches to concept-based approaches has started. In this work, it is shown by using a simple machine learning baseline, that concepts are useful as features within a machine learning framework. In all our experiments, the performance increases when including the concept-based features.