How do we determine what content to recommend on Quora?

Question

How do we determine what content to recommend on Quora?

Asked at Quora over a year ago

1.4k views

Asked at

1.4k views

How to answer Product Strategy questions

Interview Guide

Answers (3)

To answer this question, Log in.

Top Quora interview questions

How would you improve the Quora website? What will be your strategy to engage more users?2 answers | 2.6k views
Design a competitor to Quora.1 answers | 142 views
What are the top 3 metrics you would track for Quora?1 answers | 3.3k views
See Quora PM Interview Questions

Top Product Strategy interview questions

Invite members
Invite by email

Add another

Santhosh Kumar · Answer 1 · 2021-10-22T23:26:14+0000

Approach:

1. Understand the prompt better

2. Align on the business goal

3. Indentify three-five factors/dimensions that will determine the content recommendation

4. Priortize the dimensions

Clarifying questions:

Candidate: What is included in the scope of the content a) user generated b) editorial content.

Interviewer: Both

Candidate: Is there a limit on the amount of content that we recommend at a time or is it limitless

Interviewer: Limitless

Candidate: I assume we want to covers both website and app?

Interviewer: Yes

Business goal: I assume the goal here is to maximize the customer engagement with the app as measure by time spent on the app or monthly active users, etc.

Dimensions that will determine content recommendation:

We need to personalize the content recommendation per the likes and dis-likes of the individual. Therefore we need to build a profile of each individual to determine that content that we show.

Age: It is one of the key factor to ensure that we do recommend the explicit content, violent content, etc to minors.

Topics of interest: At the time of creating the account, we can ask the user to select topics that are interested in from a) politics b) sports c) lifestyle d) tech e) manufacturing f) comedy, etc. We can use this user provided input as one of the factor the make the content recommendation

Quality of content: To ensure that we recommend a quality content we need ensure that content creator has at least X follower, or it is editorial content.

Recent activity/searches: We need to generate content that closely matches with the recent topics that the user searched about or the topics that the user clicked through recently, both on the quora platform and also off the quora platform

Similar interests: Based on the historical data, we can identify the users that read content A, what other content that they also read. We can use that information to build the recommendation for the new users.

Extremists content: We need to ensure that we remove the following from the recommendation: the fake news, white supremacy related contents, conspiracy theories to promote well-being among the users and also to avoid the regulatory scrutiny.

Prioritization:

Important dimensions to base off the recommendation in that order

1. Topics of interest: I would give this a highest weightage since it is self-selected by the user

2. Recent activity/searches: this will ensure that we tailor the content according to the current disposition/state of mind of the user

3. Quality of content: this will increase the probability that the user will enjoy of the content

4. Similar interests: we can leverage the historical data to identify the content that the user would like.

5. Age/extremist conent: itis key to promote the well-being of the users and avoid the regulator scrutiny

AnshPM · Answer 2 · 2021-10-20T15:23:00+0000

How do we determine what content to recommend on Quora?

Few clarifying questions to scope the answer:

Q1) Is the content recommendation delivered through email or when a person logs in to the Quora app?

Q1) Are we looking to increase daily active/monthly active users by increasing the click thru's on the content

Q2) Is the goal to increase user engagements by providing more specific content recommendations?

I am going to make the following assumptions: content recommendation through email to increase DAU/MAU.

Quora is a Q&A platform where users can post any question and members can answer. As the answer gets upvoted it gets and ranked through an ML program, the answer gets the top spot for the question. Quora is a collaborative platform where users can edit and refine answers and add sub-threads etc. Quora thrives on fewer ads and sponsored content and is also looking at ways to incentivize creators who contribute significantly to the platform, have high voted answers and followers.

Users of the Quora platform:

1) Any user who posts random questions

2) Creator who provides answers to the questions

3) Collaborators who adds to a given answer and make it better

4) Viewers who just consume the content, read through Q&A

High-level problem statement on why the content recommendation is critical for Quora's growth:

1) Unlike other social media or collaboration platforms, Quora has fewer triggers to bring the users (viewers) back to its website

2) There is a plethora of information available on Quora and often random content is shown to the user (viewers) that doesn't engage them to spend more time browsing Quora

Now I will list down a few product improvements for Quora to improve content recommendation for the content viewers over email (those are the assumption I made in the beginning):

1) Create a user profile based on their past visits, browsing history, click thru's, etc., and identify the top 5 areas of interest for the user. send recommended content to the user through daily emails and enrich the user profile based on CTR's.

2) Bookmark Q&A that user has read in the past and create a graph on similar questions that could be of interest to them and recommend.

3) If the user has viewed nested content i.e not only viewed the top-rated answer but has also read other answers, add more weightage to these topics and recommend content to the user.

4) If the user has explicitly chosen topics that they are interested in the Quora app, then send them the top-voted answers across those topics as part of the daily recommender.

5) Crawler filter: Capture questions that users have asked on search engines or other sites and recommend them answers that have been posted on Quora relevant to their questions.

6) Add a section for recommendations from friends. When a user sees 5 posts that are recommended to them by someone they know, there is a higher chance of a user clicking through and stumbling upon a gem that they didn't know was out there. They are more likely to recommend content to other people in their network thereby triggering viral adoption.

To summarize, Quora is already providing daily recommendations to its viewers over email but to create the hook effect, they need to be crisper in their recommendations based on refining their user profiles and graphs. Adding a feature for content recommended by friends will definitely drive higher DAU and MAU for the service.

Prateek Sachdeva · Answer 3 · 2023-10-06T08:04:07+0000

Clarifying questions:

What kind of content are we referring to ? Any specific topic / sub-topic? (all)
Do we have a current recommendation strategy on Quora? If yes then what are we doi ng? (No)
Is this for a specific type of user cohort? new user vs old user? core users / casual / power? (To be prioritised) should cover all.
Is this for a specific geographic region? No
Any specific factors that should be kept in mind while recommending? No
Do we have any constraints as to what sort of content we can recommend or not? (No)
For what channel is this recommendation strategy? Quora homepage? Quora Search page? targeting via email? SEO? (It’s for the Quora homepage)

About the product/ Goal of the product:

Quora is a platform used for knowledge sharing. It thrives on user generated content and is available globally. On this platform users create questions and post them, other users on the platform answer the question. Answers then have the functionality where they can get upvoted by other readers. Users can also reply to the answers. The answers with the most number of upvotes are then put on the top spot of the recommendation algorithm.

As the consumer economy is dependent mostly on the creator economy, our goal is to boost engagement rate of the platform. To essentially increase DAU & MAU.

Users:

Creators

People who post questions on the platform

People who post answers on the platform

People who create new threads /sub-threads

People who are replying to these questions

Consumers

Silent readers of these questions / answers

Upvote answers

Goal of this recommendation algorithm is to get readers to read more and creators to create more content:

Factors that affect this phenomenon:

Globally trending topics & regionally trending topics: One of the top ways a user can land on Quora is via the SERP and through keywords. Quora being a SEO dependent platform, needs to identify what topics are trending regionally (where the user is based) and globally and accordingly curate answers/ questions for the homepage related to them. They should show answers with maximum views/ upvotes / answers etc. This is especially important for a new user who lands on the platform.
Past searched / followed topics: Recommended topics should be on the basis of the keywords that a user has searched for in the past at a high frequency or from topics/ discussions that the user is following. We can get this data by analyzing historical search pattern data of the user.
Authors that a user has interacted with the most: Recommend content from authors whose content the user has engaged with the most. It can be in the form of upvotes/ comments/ replies / answers. We should do a sentimental analysis (NLP) to gauge the sentiments asscociated with this engagement and ensure that it is a positive engagement.
Determining what friends / family are reading/ creating on Quora: Quora signups are mostly via Google accounts. We can easily capture what the user’s friends and family are reading / engaging with via their Google accounts and recommend similar content to the user. We should then track whether the user is engaging with this content or not (reading nested threads or just skimming from the top) Focusing more on the depth of enagement rather than frequency and breadth.
Tracking user’s activity on the SERP via cookies: Not limited to just Quora, we can use 3rd party cookies (with the user’s consent) and track the user’s activity (searched keywords, likes & dislikes) and accordingly recommend content to them on Quora.
Age / Behavioural sensitive content: Lastly implementing certain checks on this algorithm so as to not recommend sexual/ violent/ drug-abuse related content to children or even adults that might be triggered by such content.Age can easily be tracked but for sensitive content (we should have flags in place to track what kind of content is an adult searching for). In case a user is searching/ reading a lot of content related to physical self-harm / suicide, the user would be flagged and we would not recommend similar content on the homefeed.
Track what kind of retargeting campaigns (in this case content) works: Quora does a lot of retargeting for churned users via emails. They send specific questions/ answers on the email that might pique the user’s interest. We should also track the CTR of these questions and then the session duration post landing back on quora, Content with a high CTR and a longer followup session should be flagged and similar content should be shown on a resurrected user’s homefeed to ensure retention.