Project is about sentiment analysis of hotel reviews.
In this task firstly, the reviews are preprocessed, tokenized, lemmatized, splitted into sentences
and tagger as NER with semantics dictionary with type(pos/neg), value and multiplier.
The sentences without semantic tag are filtered out to make traverse efficient in future works.
To improve topics control lemma information is used.
Then based on user-entered topic, synonyms and hypernyms words(based on Wordnet) are generated.
Then as structural each reviews content is accepted as list of sentences. And topic based informations
are scored based on sentences which includes topics.
There are some weak points in current state:
- sarcasm like general NLP problems
- semantics dictionary elements which includes more than one word.
- Room was bad, breakfast was good.-sentence like this will be scored same and incorrect, because it calculates general score based on topic-contained sentence. ... Briefly, it is simplified task and needs to be improved.
To test the application, from root folder run shell script "mvn clean install" and then run the ReviewSentimentAnalysis/src/main/task/Main.java.
