Personalized News Recommendation based on Twitter User Modeling
1 like721 views
This document describes a personalized news recommendation system that uses different user modeling techniques. It aims to recommend news that interests individual users by understanding their behaviors and interests over time from social media data. The system collects data from social feeds, analyzes user profiles based on hashtags, entities, topics and other attributes, and uses this to build user models to power a recommendation engine. It presents the challenges of modeling millions of users and attributes at scale, and prototypes the system using a Twitter corpus to demonstrate improved news recommendations based on enriched user profiles.
![Challenge 1
• Design of User Model
[across million attributes and users]
• Design Similarity of Users
User Model](https://image.slidesharecdn.com/ire-presentation-140419045818-phpapp01/85/Personalized-News-Recommendation-based-on-Twitter-User-Modeling-5-320.jpg)
![Challenge 2
• Design of Recommendation
[filtering across million attributes and users]
• Design ranking in user attributes [topic, entity,
hash tag ]
Recommendation](https://image.slidesharecdn.com/ire-presentation-140419045818-phpapp01/85/Personalized-News-Recommendation-based-on-Twitter-User-Modeling-6-320.jpg)
![Results
• Hash tag profiles grow quickly and last for
period [Used Weka-Excel]
• Entity and Topic Profiles suggest relevant
suggestions
• Enrichment in Topics and Entity help to
correctly judge “subject”
<Improved Profile>
TEMPORAL EFFECT
PROFILE EFFECT
Enrichment EFFECT](https://image.slidesharecdn.com/ire-presentation-140419045818-phpapp01/85/Personalized-News-Recommendation-based-on-Twitter-User-Modeling-11-320.jpg)
Personalized News Recommendation based on Twitter User Modeling
- 1. Personalized News Recommendation System with Different User Modeling Engineers: Mahesh Attarde (201205685) Arpita Raj Gupta (201101121) Pankhuri Goyal (201101174)
- 2. Problem INFORMATION OVERLOAD Understand “Me” Recommend news that “Interest” Me! Use Social My “Social Behavior” Learn Me over “Time”
- 3. Problem INFORMATION OVERLOAD Understand “Me” Recommend news that “Interest” Me! Use Social My “Social Behavior” Learn Me over “Time” Personalized News Recommendation
- 4. Social Feed -> Web Tweet Processing Enrichment User Analysis User Models URL enrichment Topic/Entity Enrichment Entity/HashTag/Topic Identification Temporal Classification Personalized
- 5. Challenge 1 • Design of User Model [across million attributes and users] • Design Similarity of Users User Model
- 6. Challenge 2 • Design of Recommendation [filtering across million attributes and users] • Design ranking in user attributes [topic, entity, hash tag ] Recommendation
- 7. User Model -> Rec Engine • Temporal Feature • Profile Types • Enrichment Types
- 8. Prototype – What we had • Twitter 7.6 GB corpus • Over 1.2 years • 1218 users with avg. of 175 tweets per user • News corpus of 75,000+ articles
- 9. Prototype – What we Found • Hash Tags distinct 100000+ • Topics - distinct 18 types • Entities – distinct 39 types Temporal Feature of Hash Tag
- 10. Prototype – our Play • User Profiles 1)Hash Tag Profile 2)Entity Profile 3)Topic Profile 4)Experimental enriched • Recommendation Engine cosine similarity based
- 11. Results • Hash tag profiles grow quickly and last for period [Used Weka-Excel] • Entity and Topic Profiles suggest relevant suggestions • Enrichment in Topics and Entity help to correctly judge “subject” <Improved Profile> TEMPORAL EFFECT PROFILE EFFECT Enrichment EFFECT
- 12. Snap Shots
- 13. Snap Shots
- 14. Snap Shots
- 15. Snap Shots
- 16. Tools and Lib • Weka • Excel • Open Cailais • Stanford NPL lib • Apache Mahout • RDBMS -mysql
- 17. Thank You ! Professor Vasudev Varma Aishvarya Singh Guided By Mentored By