No abstract

The primary goal of the research reported in this paper is to identify what criteria are responsible for the good performance of a heuristic rule evaluation function in a greedy top-down covering algorithm. We first argue that search heuristics for inductive rule learning algorithms typically trade off consistency and coverage, and we investigate this trade-off by determining optimal parameter settings for five different parametrized heuristics. In order to avoid biasing our study by known functional families, we also investigate the potential of using metalearning for obtaining alternative rule learning heuristics. The key results of this experimental study are not only practical default values for commonly used heuristics and a broad comparative evaluation of known and novel rule learning heuristics, but we also gain theoretical insights into factors that are responsible for a good performance. For example, we observe that consistency should be weighted more heavily than coverage, presumably because a lack of coverage can later be corrected by learning additional rules.

Dependencies between the labels are commonly regarded as the crucial issue in multi-label classification. Rules provide a natural way for symbolically describing such relationships. For instance, rules with label tests in the body allow for representing directed dependencies like implications, subsumptions, or exclusions. Moreover, rules naturally allow to jointly capture both local and global label dependencies. In this paper, we introduce two approaches for learning such label-dependent rules. Our first solution is a bootstrapped stacking approach which can be built on top of a conventional rule learning algorithm. For this, we learn for each label a separate ruleset, but we include the remaining labels as additional attributes in the training instances. The second approach goes one step further by adapting the commonly used separate-and-conquer algorithm for learning multi-label rules. The main idea is to re-include the covered examples with the predicted labels so that this information can be used for learning subsequent rules. Both approaches allow for making label dependencies explicit in the rules. In addition, the usage of standard rule learning techniques targeted at producing accurate predictions ensures that the found rules are useful for actual classification. Our experiments show (a) that the discovered dependencies contribute to the understanding and improve the analysis of multi-label datasets, and (b) that the found multi-label rules are crucial for the predictive performance as our proposed approaches beat the baseline using conventional rules.

Social media is a rich source of up-to-date information about events such as incidents. The sheer amount of available information makes machine learning approaches a necessity to process this information further. This learning problem is often concerned with regionally restricted datasets such as data from only one city. Because social media data such as tweets varies considerably across different cities, the training of efficient models requires labeling data from each city of interest, which is costly and time consuming. To avoid such an expensive labeling procedure, a generalizable model can be trained on data from one city and then applied to data from different cities. In this paper, we present Semantic Abstraction to improve the generalization of tweet classification. In particular, we derive features from Linked Open Data and include location and temporal mentions. A comprehensive evaluation on twenty datasets from ten different cities shows that Semantic Abstraction is indeed a valuable means for improving generalization. We show that this not only holds for a two-class problem where incident-related tweets are separated from non-related ones but also for a four-class problem where three different incident types and a neutral class are distinguished. To get a thorough understanding of the generalization problem itself, we closely examined rule-based models from our evaluation. We conclude that on the one hand, the quality of the model strongly depends on the class distribution. On the other hand, the rules learned on cities with an equal class distribution are in most cases much more intuitive than those induced from skewed distributions. We also found that most of the learned rules rely on the novel semantically abstracted features.

Abstract. In this paper, we argue that search heuristics for inductive rule learning algorithms typically trade off consistency and coverage, and we investigate this trade-off by determining optimal parameter settings for five different parametrized heuristics. This empirical comparison yields several interesting results. Of considerable practical importance are the default values that we establish for these heuristics, and for which we show that they outperform commonly used instantiations of these heuristics. We also gain some theoretical insights. For example, we note that it is important to relate the rule coverage to the class distribution, but that the true positive rate should be weighted more heavily than the false positive rate. We also find that the optimal parameter settings of these heuristics effectively implement quite similar preference criteria.