---
title: "Comment by Olanrewaju Oyedele on \"Comprehensive Guide to Building Recommendation Engine\" | Hashnode"
description: "this is super informative! But hey, I feel like collaborative filtering could get kinda limited with niche items or new users. How would you tackle that?"
url: "https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32/comment/6724b8370fbce01d25f08752"
---

[Discussion](https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32)

Comment [Comprehensive Guide to Building Recommendation Engine](https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32)

[Olanrewaju Oyedele](https://hashnode.com/@olanrewaju798)

[Nov 1, 2024](https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32/comment/6724b8370fbce01d25f08752)

this is super informative! But hey, I feel like collaborative filtering could get kinda limited with niche items or new users. How would you tackle that?

[Hitesh Sanjay Nalamwar](https://hashnode.com/@HiteshNalamwar)

[Nov 1, 2024](https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32/comment/6724b8370fbce01d25f08752#reply-6724b88c096a5c859e3f5f94)

Great point! Collaborative filtering does struggle in those cases, especially with the ‘cold start’ problem. I mentioned some solutions, like using a damped mean or confidence intervals to counteract that. Do you think those would cover it, or would you look for something else?

[Olanrewaju Oyedele](https://hashnode.com/@olanrewaju798)

[Nov 1, 2024](https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32/comment/6724b8370fbce01d25f08752#reply-6724b8d66e8019ba129c47c1)

[Hitesh Sanjay Nalamwar](https://hashnode.com/@HiteshNalamwar) I see! So it’s like giving new items a ‘safe’ score until there’s enough data. I think the confidence interval thing is interesting but sounds complex. Is it something that could work for smaller sites, too?

[Hitesh Sanjay Nalamwar](https://hashnode.com/@HiteshNalamwar)

[Nov 1, 2024](https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32/comment/6724b8370fbce01d25f08752#reply-6724b923862fec23b9d000f6)

[Olanrewaju Oyedele](https://hashnode.com/@olanrewaju798) Absolutely, it can be adapted for smaller sites! You’d just need a simplified approach to calculate a baseline average score, maybe even just at the category level. That way, every new item gets a fair start without needing tons of data right away.

[Olanrewaju Oyedele](https://hashnode.com/@olanrewaju798)

[Nov 1, 2024](https://hashnode.com/posts/comprehensive-guide-to-building-recommendation-engine/66aa3eaa9a179c4902113f32/comment/6724b8370fbce01d25f08752#reply-6724b98c62bf3976734a66ef)

makes sense