Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/alinski29/opinion-miner
Mine pairs of (aspect, sentiment), with aspect topics and sentiment scores from customer reviews
https://github.com/alinski29/opinion-miner
Last synced: about 1 month ago
JSON representation
Mine pairs of (aspect, sentiment), with aspect topics and sentiment scores from customer reviews
- Host: GitHub
- URL: https://github.com/alinski29/opinion-miner
- Owner: alinski29
- License: mit
- Created: 2022-03-26T20:41:47.000Z (almost 3 years ago)
- Default Branch: master
- Last Pushed: 2022-04-18T16:29:33.000Z (over 2 years ago)
- Last Synced: 2023-05-19T13:43:19.271Z (over 1 year ago)
- Language: Scala
- Size: 24.4 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Oppinion miner
Mine pairs of (aspect, sentiment), with aspect topics and sentiment scores from customer reviews.
Table of contents
---
- [Context](#context)
- [Usage](#usage)
- [Build](#build)
- [Tweaks](#tweaks)
- [Changing word vectors](#changing-word-vectors)
- [Changing sentiment scores](changing-sentiment-scores)
- [Changing topics](changing-topics)
---
## **Context**
Traditional opinion mining extracts pairs of (aspect, sentiment) fron text by using NLP models for lemmatization, part-of-speech tagging, dependency parsing.
This work builds upon that by combining this approach with word vectors ([word2vec](https://en.wikipedia.org/wiki/Word2vec)), which are used to classify the aspects into user-defined topics by determining how similar their vectors are. Sentiment scores are assigned to sentiment words, making it possible to create topic-level aggregations.
---
## **Usage**
```scala
import com.github.alinski.opinion.miner.OpinionMiner._
val review = """
| The food was fantastic, but the service was really slow,
| although the waiter was very knowledgeable.
| The pumpkin soup was delightful, perfect for the fall season.
| Impressive rooftop and a terrific view.
| The bill was reasonable, prices were okay overall.
""".stripMargin
val opinions = OpinionMiner(review)
opinions.foreach(println)
(Aspect(waiter ∈ [service]), Sentiment(knowledgeable, 4.38))
(Aspect(service ∈ [service]), Sentiment(slow, 2.68))
(Aspect(food ∈ [food]), Sentiment(fantastic, 4.61))
(Aspect(pumpkin soup ∈ [food]), Sentiment(delightful, 4.45))
(Aspect(view ∈ [ambiance]), Sentiment(Impressive, 4.19))
(Aspect(view ∈ [ambiance]), Sentiment(terrific, 4.54))
(Aspect(rooftop ∈ [location]), Sentiment(Impressive, 4.19))
(Aspect(bill ∈ [price]), Sentiment(reasonable, 4.23))
```
See [example](https://github.com/alinski29/opinion-miner/blob/master/src/main/scala/com/github/alinski/opinion/miner/ExampleApp.scala)
---
## **Build**
Currently you have to build it from source using sbt. Should be soon available on a public repository.
```bash
sbt clean assembly
```
---
## **Tweaks**
### **Changing word vectors**
The word vectors are trained based on the [Yelp Dataset Challenge](https://www.yelp.com/dataset) 2016, on ~4 milion reviews using [FastText](https://fasttext.cc/) skipgram model. To keep the file size reasonable, only word lemmas wore kept and some infrequent words were removed.
If you want better precision or a different context, you may input your own word vectors, by providing a file with the same structure as `src/main/resources/words.vec`.
### **Changing sentiment scores**
Sentiment scores were computed for words a ngrams from ~4 mil Yelp reviews. The values are in `src/main/resources/sentimentScores.txt`. Change this file if you have different scores.
### **Changing topics**
Any number of topics can be created by modifying `src/main/resources/topicsMain.txt` / `topicsSecondary.txt`.
Topics are defined in a the following format: 1 topic per line, comma separated, first word is the topic name, the following are example members (seed words). The seed words are used to attribute a word to a topic base on their cosine similarity.
---