Sources:
- In case the link’s don’t load, there may be copyright issues with rpubs and forbes, you can copy paste these links yourself to see the sources.
Quora
Everyday I receive recommendation emails related to data science and programming from Quora. I found this Article
Scenario Design
1. Target Users?
People seeking community based answers to a wide range of topics
2 Key goals?
- Provide relevant topics for discussion
- Connect experts in certain areas to appropriate question topics
- Correctly rank answers
- Monetize-Direct ads based on topics of interest to users
3 How can I help them acheive their goals?
- It seems as though Quora is rather ahead of the curve when it comes to it’s application of Machine Learning and recommender systems. Perhaps it’s one area for growth would be in expanding its user base to other parts of the world
Extension of scenario design
- The industry level scenario design is mostly about providing a better user experience. While user’s can possibly attempt to monetize their answers, i don’t believe that is the goal of many users. Therefore, a user level scenario design is not necessary.
Application of recommender systems
- From what I have gathered these are the main topics Quora has invested time into exploring
1. User searches
- Classify questions by quality and question topic labeling
- “We use features derived from the question and its context, e.g. the user who asked the question, the locale where the question was asked…” etc." same article as before
2. Ranking answers and picking experts
- Users can ask questions labeled (A2A)- which is a request for expert answers
- To determine what an expert is, Quora uses a supervised item-wise regression approach. They conducted both online( ab testing) and offline(compare their models “good answers” to generally thought of “good answers”)
- Definition of a good answer Link to Ml process used by Quora
- Answers the question that was asked.
- Provides knowledge that is reusable by anyone interested in the question.
- Answers that are supported with rationale.
- Demonstrates credibility and is factually correct.
- Is clear and easy to read.
- Features for testing
- Text-based features
- Expertise-based features,
- Author/upvoter history-based features.
- Definition of a good answer Link to Ml process used by Quora
4. Ad optimization
- Ads CTR prediction- how often will people click on the advertisement.
- Allows businesses to target users and users to gain access to businesses they are interested in
Tools Quora uses
Directly from 1st Forbes articlek
- Logistic Regression
- Elastic Nets
- Gradient Boosted Decision Trees
- Random Forests
- (Deep) Neural Networks
- LambdaMART
- Matrix Factorization (SVD, BPR, Weighted ALS etc.)
- Vector models and other NLP techniques
- k-means and other clustering approaches
Recommendations for Quora
- Randomization for low level user
- The site has many knowledgeable individuals in the field working to improve it’s recommender systems. From my personal experience I know I have not participated in the community much and the recommender system seems to recommend rather monotonous topics. Perhaps some randomization is needed for users like myself