An interesting topic caused my contemplation about sentiment analysis: Sentiment analysis on emoji. Yingya shared one article about emoji on social media, then she posted a statement that future sentiment could analyze emoji besides text-based information. Thus, I start to imagine that if it is possible. My first thought is that emoji could somewhat represent human’s sentiment in some extend.
The emoji in row1, column 4 and 5, are pretty standard expression of love and kiss.
However, not every emoji has very unified meaning to everybody. Just like emoji in row2, column2. Based on my understanding, I often used it as helpless, neither crying nor smile. However, my roommate thinks it means laugh till tears. Emoji in row3, column2, I would like to use it when I don’t want to speak, while my roommate think it means ill. Also, emoji in row2, column7, I could either use it for surprising or frightening. Moreover, emoji in row2, column5.. I even don’t know its meaning.
Different people have different understanding towards the same emoji. Even the same person might have different use in different context. Thus, how can we use purely sentiment analysis techniques for emoji?
Another article Yingya shared on Facebook mentioned that automatic sentiment analysis could be inaccurate, because slang and sarcasm are really difficult to detect. I think it is the same as emoji. Some people might use a “smile” emoji, but how can you know it is not a smirk or irony?
In conclusion, even thought sentiment analysis has developed fast, core issue like detecting sarcasm and slang is still under solved. Emoji different from text, is even difficult to detect correctly. One possible idea I could propose might be connect the demographics with the emoji usage, also using the post-demograhics to categorize the emoji, then try to apply categorization algorithms for further analysis.
Finally, as for my research progress, unfortunately, I was pretty busy this week, so I had little time working on it. I’ve finished my survey questions, next I will post it on Mturk or other crowdsourcing websites to get the results.