Robust Compositional Polarity Classification

International Conference RANLP 2009 - Borovets, Bulgaria, pages 180–184 Robust Compositional Polarity Classification Manfred Klenner & Stefanos Petrakis & Angela Fahrni Computational Linguistics Zurich University, Switzerland {klenner,petrakis}@cl.uzh.ch angela.fahrni@swissonline.ch Abstract We describe a pattern-based system for polarity classification from texts. Our system is currently restricted to the positive, negative or neutral po- larity of phrases and sentences. It analyses the input texts with the aid of a polarity lexicon that specifies the prior polarity of words. A chunker is used to determine phrases that are the basis for a compositional treatment of phrase-level po- larity assignment. In our current experiments we focus on sentences that are targeted towards per- sons, be it the writer (I, my, me, ..), the social group including the writer (we, our, ..) or the reader (you, your, ..). We evaluate our system on a manually annotated set of sentences taken from texts from a panel group called ’I battle depression’. We present the results of comparing our system’s performance over this gold standard against a baseline system. Keywords sentiment analysis, polarity composition 1 Introduction Polarity classification aims at identifying the positive and negative polarities of text, at various levels, in- cluding document, sentence, phrase and word level. Such a task can be guided by the principle of compositionality[5], which states that: ”The meaning of a complex expression is de- termined by its structure and the meanings of its constituents.” Based on this principle, the polarity of a portion of text can be composed from the polarities of its con- stituents in a systematic way[11]. An example of such compositionality appears in the following sentence, ”He is a good liar” which is classified as negative1, since a positive ad- jective and a negative noun yield a negative noun phrase. In principle, such an incremental composi- tional interpretation might proceed up the sentence level — negating, confirming and intensifying already computed phrase polarities. In the sentence: ”He is a quite good liar.” 1 With an overtone of admiration the positive polarity of the adjective is confirmed and intensified (’quite’), whereas in the sentence ”He is not an extremely good liar.” the positive polarity of the adjective remains but is de- creased. This happens because the adverb ’extremely’ is shifted by the ”not” negator and does not function as an intensifier anymore. Negation is the most common form of so-called polarity shifters. Another example is ’without’ - ’without hope’ is negative, but ’without fear’ is positive. In the simplest case, word polarities are provided by a polarity lexicon. Commonly used lexicons are the subjectivity lexicon from [15], the semi-automatically derived SentiWordNet [6] or lexicons generated from the General Inquirer lexicon [12]. Ambiguity turns out to be a problem: ’a cheap ther- apy’ might be regarded as positive if ’cheap’ means ’low price’ but negative if it means ’low quality’. How- ever, we have identified only few cases of ambiguity in our experiments. In principle, we identify ambiguity of this type as a challenging problem, although we don’t cope with …

Content truncated to 500 words