In Japan, a bank analyst is asked to prepare a report comparing attitudes toward mobile banking in Japan and Indonesia. The analyst has a rich resource of text for information - comments collected through Japanese and Indonesian banking websites.  But the analyst speaks only Japanese. A financial analyst in Thailand needs to know whether the posts about imported rice by Facebook users in Malaysia are predominantly positive or negative.  An investor in London wonders whether opinions expressed on linkhay, the Vietnamese equivalent of twitter, could be used to predict Vietnamese markets.  Around the world, business people are faced with their own versions of the same problem: they all need information that is embedded in unfamiliar languages.
 
Cross-lingual sentiment analysis is the practice of assessing attitudes expressed in text that is foreign to the researcher. Consider the example of the Thai financial analyst mentioned earlier. A Thai speaker requires information found in Malay text; that’s cross-lingual. The information required is a summary of the attitudes expressed in the text – based on positive/neutral negative categories, or numeric scores; that’s sentiment analysis.
 
The analyst would face one obvious challenge – the inability to read the Malay language – as well as several which are less obvious. Consider the major steps involved: identify a topic of interest, retrieve the relevant text, classify the sentiment of each document and finally, summarize the results. These are not simple tasks even when only one language is involved. In a cross-lingual context, the process is even more complex.
 
Here are some issues to consider when planning a cross-lingual sentiment analysis project:
 
The topic must be defined fully and clearly.
 
Beginning with the researcher’s native language, a list of relevant search terms must be prepared. Perhaps the researcher is interested in attitudes tow...

Categories:

Data & Analytics,Data Management,Risk & Performance,

Data & Analytics,Data Management,Risk & Performance,

Keywords:Cross Lingual Sentiment Analysis, Social Media, Native Language