Alexander Hogenboom

Exploiting emoticons in sentiment analysis

As people increasingly use emoticons in text in order to express, stress, or disambiguate their sentiment, it is crucial for automated sentiment analysis tools to correctly account for such graphical cues for sentiment. We analyze how emoticons typically convey sentiment and demonstrate how we can exploit this by using a novel, manually created emoticon sentiment lexicon in order to improve a state-of-the-art lexicon-based sentiment classification method. We evaluate our approach on 2,080 Dutch tweets and forum messages, which all contain emoticons and have been manually annotated for sentiment. On this corpus, paragraph-level accounting for sentiment implied by emoticons significantly improves sentiment classification accuracy. This indicates that whenever emoticons are used, their associated sentiment dominates the sentiment conveyed by textual cues and forms a good proxy for intended sentiment.

Semantics-based information extraction for detecting economic events

As today’s financial markets are sensitive to breaking news on economic events, accurate and timely automatic identification of events in news items is crucial. Unstructured news items originating from many heterogeneous sources have to be mined in order to extract knowledge useful for guiding decision making processes. Hence, we propose the Semantics-Based Pipeline for Economic Event Detection (SPEED), focusing on extracting financial events from news articles and annotating these with meta-data at a speed that enables real-time use. In our implementation, we use some components of an existing framework as well as new components, e.g., a high-performance Ontology Gazetteer, a Word Group Look-Up component, a Word Sense Disambiguator, and components for detecting economic events. Through their interaction with a domain-specific ontology, our novel, semantically enabled components constitute a feedback loop which fosters future reuse of acquired knowledge in the event detection process.

Multi-lingual support for lexicon-based sentiment analysis guided by semantics

Many sentiment analysis methods rely on sentiment lexicons, containing words and their associated sentiment, and are tailored to one specific language. Yet, the ever-growing amount of data in different languages on the Web renders multi-lingual support increasingly important. In this paper, we assess various methods for supporting an additional target language in lexicon-based sentiment analysis. As a baseline, we automatically translate text into a reference language for which a sentiment lexicon is available, and subsequently analyze the translated text. Second, we consider mapping sentiment scores from a semantically enabled sentiment lexicon in the reference language to a new target sentiment lexicon, by traversing relations between language-specific semantic lexicons. Last, we consider creating a target sentiment lexicon by propagating sentiment of seed words in a semantic lexicon for the target language. When extending sentiment analysis from English to Dutch, mapping sentiment across languages by exploiting relations between semantic lexicons yields a significant performance improvement over the baseline of about 29% in terms of accuracy and macro-level F1 on our data. Propagating sentiment in language-specific semantic lexicons can outperform the baseline by up to about 47%, depending on the seed set of sentiment-carrying words. This indicates that sentiment is not only linked to word meanings, but tends to have a language-specific dimension as well.

Sentiment Lexicon Creation from Lexical Resources

Today’s business information systems face the challenge of analyzing sentiment in massive data sets for supporting, e.g., reputation management. Many approaches rely on lexical resources containing words and their associated sentiment. We perform a corpus-based evaluation of several automated methods for creating such lexicons, exploiting vast lexical resources. We consider propagating the sentiment of a seed set of words through semantic relations or through PageRank-based similarities. We also consider a machine learning approach using an ensemble of classifiers. The latter approach turns out to outperform the others. However, PageRank-based propagation appears to yield a more robust sentiment classifier.

Semantic news recommendation using wordnet and bing similarities

While traditionally content-based news recommendation was performed using the word vector space model, more recent approaches also take into account semantics, often through the use of semantic lexicons. However, named entities are rarely taken into account, as they are often absent in such lexicons. Nevertheless, they can play a crucial role in determining user interest for specific news articles. Therefore, in this work, we extend the state-of-the-art semantic lexicon-driven Semantic Similarity (SS) recommendation method by additionally considering named entities. First, as in SS, we calculate similarities between WordNet synonym sets in unread news items and synonym sets in read news items (stored in user profiles). Then, we use the page counts of named entities that are retrieved from the Bing Web search engine to compute named entity similarities between unread and read news items. Results show that our recommendation method, BingSS, outperforms SS in terms of F1, precision, accuracy, and specificity.

Analyzing Sentiment in a Large Set of Web Data While Accounting for Negation

As virtual utterances of opinions or sentiment are becoming increasingly abundant on the Web, automated ways of analyzing sentiment in such data are becoming more and more urgent. In this paper, we provide a classification scheme for existing approaches to document sentiment analysis. As the role of negations in sentiment analysis has been explored only to a limited extent, we additionally investigate the impact of taking into account negation when analyzing sentiment. To this end, we utilize a basic sentiment analysis framework - consisting of a wordbank creation part and a document scoring part - taking into account negation. Our experimental results show that by accounting for negation, precision on human ratings increases with 1.17%. On a subset of selected documents containing negated words, precision increases with 2.23%.

RCQ-GA: RDF Chain Query Optimization Using Genetic Algorithms

The application of Semantic Web technologies in an Electronic Commerce environment implies a need for good support tools. Fast query engines are needed for efficient querying of large amounts of data, usually represented using RDF. We focus on optimizing a special class of SPARQL queries, the so-called RDF chain queries. For this purpose, we devise a genetic algorithm called RCQ-GA that determines the order in which joins need to be performed for an efficient evaluation of RDF chain queries. The approach is benchmarked against a two-phase optimization algorithm, previously proposed in literature. The more complex a query is, the more RCQ-GA outperforms the benchmark in solution quality, execution time needed, and consistency of solution quality. When the algorithms are constrained by a time limit, the overall performance of RCQ-GA compared to the benchmark further improves.

Sentiment analysis with a multilingual pipeline

Sentiment analysis refers to retrieving an authors sentiment from a text. We analyze the differences that occur in sentiment scoring across languages. We present our experiments for the Dutch and English language based on forum, blog, news and social media texts available on the Web, where we focus on the differences in the use of a language and the effect of the grammar of a language on sentiment analysis. We propose a multilingual pipeline for evaluating how an authors sentiment is conveyed in different languages. We succeed in correctly classifying positive and negative texts with an accuracy of approximately 71% for English and 79% for Dutch. The evaluation of the results shows however that usage of common expressions, emoticons, slang language, irony, sarcasm, and cynicism, acronyms and different ways of negation in English prevent the underlying sentiment scores from being directly comparable.

Mining economic sentiment using argumentation structures

The recent turmoil in the financial markets has demonstrated the growing need for automated information monitoring tools that can help to identify the issues and patterns that matter and that can track and predict emerging events in business and economic processes. One of the techniques that can address this need is sentiment mining. Existing approaches enable the analysis of a large number of text documents, mainly based on their statistical properties and possibly combined with numeric data. Most approaches are limited to simple word counts and largely ignore semantic and structural aspects of content. Yet, argumentation plays an important role in expressing and promoting an opinion. Therefore, we propose a framework that allows the incorporation of information on argumentation structure in the models for economic sentiment discovery in text.

Ant colony optimization for RDF chain queries for decision support

Semantic Web technologies can be utilized in expert systems for decision support, allowing a user to explore in the decision making process numerous interconnected sources of data, commonly represented by means of the Resource Description Framework (RDF). In order to disclose the ever-growing amount of widely distributed RDF data to demanding users in real-time environments, fast RDF query engines are of paramount importance. A crucial task of such engines is to optimize the order in which partial results of a query are joined. Several soft computing techniques have already been proposed to address this problem, i.e., two-phase optimization (2PO) and a genetic algorithm (GA). We propose an alternative approach - an ant colony optimization (ACO) algorithm, which may be more suitable for a Semantic Web environment. Experimental results with respect to the optimization of RDF chain queries on a large RDF data source demonstrate that our approach outperforms both 2PO and a GA in terms of execution time and solution quality for queries consisting of up to 15 joins. For larger queries, both ACO and a GA may be preferable over 2PO, subject to a trade-off between execution time and solution quality. The GA yields relatively good solutions in a comparably short time frame, whereas ACO needs more time to converge to high-quality solutions.