How do you feel, señorita?
Problem being addressed
A major part of sentiment analysis research has been monolingual and best of the monolingual methods perform poorly on code-mixed text, making them inefficient in multilingual societies.
Analysis of code-mixed text is hard due to lack of code-mixed text datasets, large amount of unseen constructions caused by combining lexicon and syntax of two or more language, and large number of possible combinations of code mixed languages. Constructing one dataset requires lot of manual labour and construction of datasets for each code-mixed pair seems infeasible. The suggested methods use different kinds of multilingual and crosslingual embeddings to efficiently transfer knowledge from monolingual text to code-mixed text for sentiment analysis of code-mixed text.
Advantages of this solution
The novel methods provide an effective way to improve the performance of sentiment analysis techniques on code-mixed text. They are independent of any kind supervised signals between two languages by not using any kind of Machine Translation or Language Identification systems, as these systems perform poorly on low-resource language pair due to lack of proper datasets.
Possible New Application of the Work
Marketing industry can benefit from improved sentiment analysis as it provides better understanding of the customer feedback, especially in case of the marketplaces where the feedback can be given in mixed languages due to the customer's background and the specifics of product that's being reviewed..
Sentiment analysis is one of the most commonly used techniques to monitor social media, and in case of postings that use more than one language (e.g. in the US, where English and Spanish can be often mixed in one sentence) the suggested methodology can significantly improve the accuracy of such analysis.
Source URL: #############