Abstract—Stock trend prediction based on text has gained
much attention from researchers in recent years. According to
investment theories, investors’ behaviors will influence the
stock market, and the way people invest their money is based
on the history trend and information they hold. On account of
this indirectly influential relationship between information of
stock and stock trend, stock trend prediction based on text has
been done by many researchers. However, due to the serious
feature sparse problem in tweets and unreliability of using
average sentiment score to indicate one day’s sentiment, this
work proposed a text-sentiment based stock trend prediction
model with a hybrid feature selection method. Instead of
applying sentiment analysis to add sentiment related features,
this paper uses SentiWordNet to give an additional weight to
the selected features. Besides, this work also compares the
results with those of other learning algorithms. SVM linear
algorithm based on leave-one-out cross validation yields the
best performance of 90.34%.