ECNU: Leveraging on Ensemble of Heterogeneous Features and Information Enrichment for Cross Level Semantic Similarity Estimation

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 265–270, Dublin, Ireland, August 23-24, 2014. ECNU: Leveraging on Ensemble of Heterogeneous Features and Information Enrichment for Cross Level Semantic Similarity Estimation Tian Tian Zhu Department of Computer Science and Technology East China Normal University 51111201046@ecnu.cn Man Lan∗ Department of Computer Science and Technology East China Normal University mlan@cs.ecnu.edu.cn∗ Abstract This paper reports our submissions to the Cross Level Semantic Similarity (CLSS) task in SemEval 2014. We submitted one Random Forest regression system on each cross level text pair, i.e., Paragraph to Sentence (P-S), Sentence to Phrase (S- Ph), Phrase to Word (Ph-W) and Word to Sense (W-Se). For text pairs on P-S level and S-Ph level, we consider them as sentences and extract heterogeneous types of similarity features, i.e., string features, knowledge based features, corpus based features, syntactic features, machine trans- lation based features, multi-level text fea- tures, etc. For text pairs on Ph-W level and W-Se level, due to lack of informa- tion, most of these features are not ap- plicable or available. To overcome this problem, we propose several information enrichment methods using WordNet syn- onym and definition. Our systems rank the 2nd out of 18 teams both on Pearson cor- relation (official rank) and Spearman rank correlation. Specifically, our systems take the second place on P-S level, S-Ph level and Ph-W level and the 4th place on W-Se level in terms of Pearson correlation. 1 Introduction Semantic similarity is an essential component of many applications in Natural Language Process- ing (NLP). Previous works often focus on text se- mantic similarity on the same level, i.e., paragraph to paragraph or sentence to sentence, and many ef- fective text semantic measurements have been pro- posed (Islam and Inkpen, 2008), (B¨ar et al., 2012), This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ (Heilman and Madnani, 2012). However, in many real world cases, the two texts may not always be on the same level. The Cross Level Semantic Similarity (CLSS) task in SemEval 2014 provides a universal platform to measure the degree of se- mantic equivalence between two texts across dif- ferent levels. For each text pair on four cross lev- els, i.e., Paragraph to Sentence (P-S), Sentence to Phrase (S-Ph), Phrase to Word (Ph-W) and Word to Sense (W-Se), participants are required to re- turn a similarity score which ranges from 0 (no relation) to 4 (semantic equivalence). We partici- pate in all the four cross levels and take the second place out of all 18 teams both on Pearson correla- tion (official) and Spearman correlation ranks. In this work, we present a supervised regres- sion system for each cross level separately. For P-S level and S-Ph level, we regard the paragraph of P-S as a long sentence, and the phrase of S- Ph as a short sentence. Then we use various types of text similarity features including …

Content truncated to 500 words