Enrique Amigó

A comparison of extrinsic clustering evaluation metrics based on formal constraints

There is a wide set of evaluation metrics available to compare the quality of text clustering algorithms. In this article, we define a few intuitive formal constraints on such metrics which shed light on which aspects of the quality of a clustering are captured by different metric families. These formal constraints are validated in an experiment involving human assessments, and compared with other constraints proposed in the literature. Our analysis of a wide range of metrics shows that only BCubed satisfies all formal constraints. We also extend the analysis to the problem of overlapping clustering, where items can simultaneously belong to more than one cluster. As Bcubed cannot be directly applied to this task, we propose a modified version of Bcubed that avoids the problems found with other metrics.

Overview of RepLab 2013: Evaluating Online Reputation Monitoring Systems

This paper summarizes the goals, organization, and results of the second RepLab competitive evaluation campaign for Online Reputation Management Systems RepLab 2013. RepLab focused on the process of monitoring the reputation of companies and individuals, and asked participant systems to annotate different types of information on tweets containing the names of several companies: first tweets had to be classified as related or unrelated to the entity; relevant tweets had to be classified according to their polarity for reputation Does the content of the tweet have positive or negative implications for the reputation of the entity?, clustered in coherent topics, and clusters had to be ranked according to their priority potential reputation problems had to come first. The gold standard consists of more than 140,000 tweets annotated by a group of trained annotators supervised and monitored by reputation experts.

A general evaluation measure for document organization tasks

A number of key Information Access tasks -- Document Retrieval, Clustering, Filtering, and their combinations -- can be seen as instances of a generic {\em document organization} problem that establishes priority and relatedness relationships between documents (in other words, a problem of forming and ranking clusters). As far as we know, no analysis has been made yet on the evaluation of these tasks from a global perspective. In this paper we propose two complementary evaluation measures -- Reliability and Sensitivity -- for the generic Document Organization task which are derived from a proposed set of formal constraints (properties that any suitable measure must satisfy). In addition to be the first measures that can be applied to any mixture of ranking, clustering and filtering tasks, Reliability and Sensitivity satisfy more formal constraints than previously existing evaluation metrics for each of the subsumed tasks. Besides their formal properties, its most salient feature from an empirical point of view is their strictness: a high score according to the harmonic mean of Reliability and Sensitivity ensures a high score with any of the most popular evaluation metrics in all the Document Retrieval, Clustering and Filtering datasets used in our experiments.

Towards real-time summarization of scheduled events from twitter streams

We deal with shrinking the stream of tweets for scheduled events in real-time, following two steps: (i) sub-event detection, which determines if something new has occurred, and (ii) tweet selection, which picks a tweet to describe each sub-event. By comparing summaries in three languages to live reports by journalists, we show that simple text analysis methods which do not involve external knowledge lead to summaries that cover 84% of the sub-events on average, and 100% of key types of sub-events (such as goals in soccer).

The role of named entities in Web People Search

The ambiguity of person names in the Web has become a new area of interest for NLP researchers. This challenging problem has been formulated as the task of clustering Web search results (returned in response to a person name query) according to the individual they mention. In this paper we compare the coverage, reliability and independence of a number of features that are potential information sources for this clustering task, paying special attention to the role of named entities in the texts to be clustered. Although named entities are used in most approaches, our results show that, independently of the Machine Learning or Clustering algorithm used, named entity recognition and classification per se only make a small contribution to solve the problem.

Overview of RepLab 2014: Author Profiling and Reputation Dimensions for Online Reputation Management

This paper describes the organisation and results of RepLab 2014, the third competitive evaluation campaign for Online Reputation Management systems. This year the focus lied on two new tasks: reputation dimensions classification and author profiling, which complement the aspects of reputation analysis studied in the previous campaigns. The participants were asked (1) to classify tweets applying a standard typology of reputation dimensions and (2) categorise Twitter profiles by type of author as well as rank them according to their influence. New data collections were provided for the development and evaluation of systems that participated in this benchmarking activity.

QARLA: A Framework for the Evaluation of Text Summarization Systems

This paper presents a probabilistic framework, QARLA, for the evaluation of text summarisation systems. The input of the framework is a set of manual (reference) summaries, a set of baseline (automatic) summaries and a set of similarity metrics between summaries. It provides i) a measure to evaluate the quality of any set of similarity metrics, ii) a measure to evaluate the quality of a summary using an optimal set of similarity metrics, and iii) a measure to evaluate whether the set of baseline summaries is reliable or may produce biased results.Compared to previous approaches, our framework is able to combine different metrics and evaluate the quality of a set of metrics without any a-priori weighting of their relative importance. We provide quantitative evidence about the effectiveness of the approach to improve the automatic evaluation of text summarisation systems by combining several similarity metrics.

MT Evaluation: Human-Like vs. Human Acceptable

We present a comparative study on Machine Translation Evaluation according to two different criteria: Human Likeness and Human Acceptability. We provide empirical evidence that there is a relationship between these two kinds of evaluation: Human Likeness implies Human Acceptability but the reverse is not true. From the point of view of automatic evaluation this implies that metrics based on Human Likeness are more reliable for system tuning. Our results also show that current evaluation metrics are not always able to distinguish between automatic and human translations. In order to improve the descriptive power of current metrics we propose the use of additional syntax-based metrics, and metric combinations inside the QARLA Framework.

Learning similarity functions for topic detection in online reputation monitoring

Reputation management experts have to monitor--among others--Twitter constantly and decide, at any given time, what is being said about the entity of interest (a company, organization, personality...). Solving this reputation monitoring problem automatically as a topic detection task is both essential--manual processing of data is either costly or prohibitive--and challenging--topics of interest for reputation monitoring are usually fine-grained and suffer from data sparsity. We focus on a solution for the problem that (i) learns a pairwise tweet similarity function from previously annotated data, using all kinds of content-based and Twitter-based features; (ii) applies a clustering algorithm on the previously learned similarity function. Our experiments indicate that (i) Twitter signals can be used to improve the topic detection process with respect to using content signals only; (ii) learning a similarity function is a flexible and efficient way of introducing supervision in the topic detection clustering process. The performance of our best system is substantially better than state-of-the-art approaches and gets close to the inter-annotator agreement rate. A detailed qualitative inspection of the data further reveals two types of topics detected by reputation experts: reputation alerts / issues (which usually spike in time) and organizational topics (which are usually stable across time).

Discovering filter keywords for company name disambiguation in twitter

A major problem in monitoring the online reputation of companies, brands, and other entities is that entity names are often ambiguous (apple may refer to the company, the fruit, the singer, etc.). The problem is particularly hard in microblogging services such as Twitter, where texts are very short and there is little context to disambiguate. In this paper we address the filtering task of determining, out of a set of tweets that contain a company name, which ones do refer to the company. Our approach relies on the identification of filter keywords: those whose presence in a tweet reliably confirm (positive keywords) or discard (negative keywords) that the tweet refers to the company. We describe an algorithm to extract filter keywords that does not use any previously annotated data about the target company. The algorithm allows to classify 58% of the tweets with 75% accuracy; and those can be used to feed a machine learning algorithm to obtain a complete classification of all tweets with an overall accuracy of 73%. In comparison, a 10-fold validation of the same machine learning algorithm provides an accuracy of 85%, i.e., our unsupervised algorithm has a 14% loss with respect to its supervised counterpart. Our study also shows that (i) filter keywords for Twitter does not directly derive from the public information about the company in the Web: a manual selection of keywords from relevant web sources only covers 15% of the tweets with 86% accuracy; (ii) filter keywords can indeed be a productive way of classifying tweets: the five best possible keywords cover, in average, 28% of the tweets for a company in our test collection.