Instapy: [Feature Request] Filtering - Sentiment analysis

[timgrossmann]

Filters: this option lets you make filter people by a set of possible conditions.

Implement a way to check if a post is negative/positive, e.g. Only like posts that have a positive "way of writing" => prevent InstaPy from commenting "Awesome" on some post that features a problematic situation or some person that has an illness.

Could also be a nice idea to introduce another layer of comments which are specifically targeted at posts that have bad content like: "Get well soon!"

[Naramsim]

NLP

[gitpatrickhub]

@timgrossmann @uluQulu here are some options for sentiment analysis APIs;

https://medium.com/@sifium/top-five-emotional-sentiment-analysis-apis-116cd8d42055

Some options are free with limits like Clarifai, with ability to use a simple classification of Positive/Negative or break it down by emotions (5-8 classes).

Another option would be to use NLTK, in combination with a datasource like WordNet and provide a method of running locally. This approach would likely cause a performance hit.

Let me know your thoughts.

[uluQulu]

It is fantastic @gitpatrickhub!
It has always been in my mind to have such kind of feature but it is very difficult to provide a good behaving capability.
Cos it is prone to misbehave out of wrong return result.

Although there are great tools, it must be so important to choose the one with the highest rate of accuracy in results rather than those with multiple functionalities.

More importantly, @gitpatrickhub the question is could you implement it?

Having seen your latest work, I can say you can 🙌

[gitpatrickhub]

@uluQulu Thank you, I can definitely try. It would be best if @timgrossmann selects the preferred method for the project and I will take a stab at implementing it. May need some guidance along the way, which has been great and much appreciated.

[timgrossmann]

@uluQulu @gitpatrickhub So awesome that you're willing to take on this feature!
Like @uluQulu already mentioned I'm also more than confident that you're able to come up with and implement it.

The question about how we should go about it is really important.
I think @uluQulu already mentioned it, it's absolutely critical to make sure this feature is as "good" as possible. This means that it most likely will not be good to simply use a library or any other custom implementation which definitely won't give near the accuracy one of the online services provide...
So I think using something like clarifai for this would be the right approach. Again making sure that people are in no way forced to use it and making it as simple as possible while still keeping the complexity of allowing the user to "react" on different moods differently.

@gitpatrickhub I think Instagram tweets are more special types of text so it might be good to compare some of the services with a few example texts from Instagram before deciding. (Also very important is the handling of the emojis/encoding etc)
Would you mind doing a few quick tests with your favourites of the article you posted?
Would love to hear/read about it here. (This might also be an interesting part for the InstaPy-Research repository, adding some more than simply the API-"Research" there)

Thank you very much guys, looking forward to this!
Best
Tim

[gitpatrickhub]

@timgrossmann @uluQulu Thank you both for the input and the vote of confidence! I'll start running tests with a couple of different options and will post my results here. Excellent point regarding the emojis/encoding. I had not yet considered that aspect, but I agree it will be critical for an effective solution.