Introduction: For carrying out Amharic sentiment classification, the availability of sentiment lexicons is crucial. To date, there are two generated Amharic sentiment lexicons. These are manually generated lexicon (1000) [2] and dictionary based Amharic SWN and SOCAL lexicons [3]. However, dictionary based generated lexicons has short-comings in that it has difficulty in capturing cultural connotation and language specific features of the language. This research builds corpus based algorithm to handle language and culture specific words in the lexicons [1]. However, it could probably be impossible to handle all the words in the language as the corpus is a limited re source in almost all less resourced languages like Amharic. But still it is possible to build sentiment lexicons in particular domain where large amount of Amharic corpus is available. Due to this reason, the lexicon built using this approach is usually used for lexicon based sentiment analysis in the same domain from which it is built. The research questions to be addressed utilizing this approach are: (1) how can we build an approach to generate Amharic sentiment lexicon from corpus? (2) how do we evaluate the validity and quality of the generated lexicon? Related work: Our work is closely associated to the work of [4] which generated emotion based lexicon by bootstrapping corpus using word distributional semantics (i.e. using Positive Point-wise Mutual Information (PPMI)). Our approach is different from [4] in that we generated sentiment lexicon rather than emotion lexicon. The other thing is that the approach of propagating sentiment to expand the seeds is also different. Besides, the threshold selection, the seed words' part of speech are different from language to language. For example, Amharic has few adverb classes unlike Italian [5]. Thus, our seed words do not contain adverbs. Proposed corpus based approaches: There are variety of corpus based strategies that include count based (e.g. PPMI) and predictive based (e.g. word embedding) approaches. In this part, we present the proposed count based approach to generate Amharic sentiment lexicon from a corpus. The proposed framework of corpus based approach tries to generate Amharic sentiment lexicon. The framework has four com ponents: (Amharic news) corpus collections, preprocessing module, PPMI matrix of word-context, algorithm to generate (Amharic) sentiment lexicon resulting in the generated (Amharic) sentiment lexicon. See the framework in Fig.1 of Appendix. We developed algorithms for constructing Amharic sentiment lexicons automatically from Amharic news corpus. Corpus based approach is proposed relying on the word co-occurrence distributional embedding including frequency based embedding (i.e. PPMI). First we build word-context unigram frequency count matrix and transform it to point-wise mutual Information matrix. For an experimentally chosen threshold
value, the top closest words to the mean vector of seed list are added to the lexicon. Then, the mean vector of the new sentiment seed list is updated and process is repeated until we get sufficient terms in the lexicon.
Results: Seed words of size 519 are used to expand PPMI based lexicons. With ex
perimentally obtained threshold value of 100 and 200, we got corpus based Amharic
sentiment lexicons of size 1811 and 3794 respectively. See sample of generated
lexicon in Table 2 of Appendix. As discussed on dictionary based lexicons in [
Evaluation: We evaluated the generated Amharic sentiment lexicon in two ways: external to lexicon and internal to lexicon. External to lexicon is to test the usefulness and the correctness of each of the lexicon to find sentiment score of sentiment labeled Amharic comments corpus. Internal evaluation is compute the degree to which each of the generated lexicons are overlapped (or agreed) with manual, SOCAL and SWN (Amharic) sentiment lexicons. Our lexicon detects subjectivity of Amharic facebook comments has shown an increment of 3.73 more than the subjectivity detection rate of the manual lexicon. For sentiment classification, the performance of our generated lexicon for classifying sentiment of Amharic facebook comments has an increment of 6.71 than the manual sentiment lexicon. See evaluation of our lexicon in Table 1 of Appendix. In addition, the coverage result in a general corpus of 20 million tokens depicts that the coverage of PPMI based Amharic sentiment lexicon is better than the manual lexicon and SOCAL. However, it has less coverage than SWN. Unlike SWN, PPMI based lexicon is generated from corpus. Due to this reason its coverage to work on a general domain is limited. It also demonstrated that the positive and negative count in almost all lexicons seems to have balanced and uniform distribution of sentiment polarity terms in the corpus.
Conclusions: This study revealed that it is possible to create sentiment lexicon for low resourced languages from corpus. This captures the language specific features and connotations related to the culture where the language is spoken. This cannot be handled using dictionary based approach that propagates labels from resource rich languages. To the best of our knowledge, the PPMI based approach to generate Amharic sentiment lexicon form corpus is performed for first time for Amharic language with minimal costs and time. Thus, the generated lexicons can be used in combination with other sentiment lexicons to enhance the performance of sentiment classifications in Amharic language. The approach is a generic approach which can be adapted to other resource limited languages to reduce cost of human annotation and the time it takes to annotated sentiment lexicons. Though the PPMI based Amharic sentiment lexicon outperforms the manual lexicon, prediction (word embedding) based approach is recommended to generate sentiment lexicon for Amharic language to handle context sensitive terms.
Appendix: List of figures and tables