search

dvorka / mindforger

Thinking notebook and Markdown editor with LLM wingman.
https://www.mindforger.com
GNU General Public License v2.0
2.21k stars 133 forks source link

AI: Recommender #253

Open dvorka opened 6 years ago

dvorka commented 6 years ago

Bag of Words + doc term matrix + tf-idf ... to be implemented to AI.

See also TensorFlow word2vec. RNN (https://machinelearnings.co/tensorflow-text-classification-615198df9231).

dvorka commented 6 years ago

For now I have implemented associations in a different way, however, this is wort to review and incorporate interesting ideas later.

gekaremi commented 6 years ago

maybe as example

http://arxivist.com/ The arxivist uses your preferences to sort arXiv articles --- making it easier to find new arXiv submissions that are pertinent to you.

How do rankings work?

Loosley speaking, the arxivist skims the title, authors and categories of each article to see whether it matches up with your interests as indicated through upvotes/downvotes of previous articles.

More specifically, the arxivist keeps track of keywords for each article: the authors and categories, as well as words and phrases in the title. When you upvote/downvote an article, the arxivist interprets that as a upvote/downvote of each keyword for that article. When the arxivist needs to rate a new article for you, it looks at your preference for each of the article's keywords and sums them up to get the total score.

For example, if you click thumbs-up for 10 articles in the category "math.MG", then any article in the category "math.MG" will gain 10 points towards its computed score. Likewise, if you upvote 10 articles with the word "hyperbolic" in the title, any other article with the word "hyperbolic" will gain 10 points. Because word pairs are keywords as well, phrases like "Heisenberg group" count more strongly: liking 10 articles with the phrase "Heisenberg group" will give add a score of 30 towards any article with that phrase in the title.

http://paperscape.org/ A map of 1,407,463 scientific papers from the arXiv. Last updated: 4 July 2018

dvorka commented 6 years ago

This is great and relevant! I just signed up and looking for experimenting with/learning arxivist, ... and http://paperscape.org visualization is astonishing :wink:

gekaremi commented 6 years ago

http://utopiadocs.com/lazarus/

Utopia Documents is provided free under an open source licence (GPL version 3).

Follow expressions Lazarus reads your articles alongside you and finds what it thinks are import expressions – fragments of sentences that represent the ideas portrayed in the literature. Whether regarding the activation of a particular gene, or a drug's therapeutic effect, these expressions can be used to find new articles that talk about similar concepts.

Explore meaning Lazarus provides a whole new set of sources for exploring the meaning of terms and phrases you encounter while reading. Any text that has a slight red tint can be explored with a single click, providing links to databases and new definitions.

Enrich articles Lazarus also pulls out section headings and bibliographies from those articles for which Utopia has been unable to find online information, allowing you to easily traverse citations in more literature than ever before.

http://www.cs.manchester.ac.uk/our-research/activities/lazarus/

dvorka commented 6 years ago

Lazarus is very interesting project + it's definitely in the direction I would like to go. It seems to be complex and comprehensive. The second link you shared is even more valuable as it gives overview of how Lazarus has been built.

@gekaremi thank you for making this comprehensive research for me! I really appreciate it.

gekaremi commented 6 years ago

@dvorka Thank you! I am glad to help