Predicting the Popularity of Instagram Posts for a Lifestyle Mag

展开查看详情

1.Predicting the Popularity of Instagram Posts for a Lifestyle Magazine Using Deep Learning Shaunak De, Abhishek Maity, Vritti Goel, Sanjay Shitole and Avik Bhattacharya

2. Introduction ● Instagram is a social media platform for visual media-sharing. ● Increasingly being adopted by traditional media platforms like magazines. ● Analysis of the popularity or traction of the Instagram posts becomes important for estimation of reach etc. ● In a commercial scenario it is important to be able to coarsely predict the reach of a particular post for price fixation with advertisers. "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

3. Commercial Interest ● Reach Analysis ○ Estimation of reader interaction ○ Influence of magazine ○ Brand Image ● Advertorial Price Fixing ○ Pricing depends on reader interaction ○ Enforceable and measurable impact "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

4. Typical Post "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

5. Metrics "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

6. Challenge of “Tag” discovery ● #watches is related to a post tagged #seiko ● Because ‘Seiko‘ is a manufacturer of ‘watches’. ● However, lexicographically they have little inter-relation ● Solution: The use of a word-tree ○ Post A contains the tags #watch, #cricket and #sachin ○ Post B contains the tags #cricket, #game, ○ Then both posts are to be grouped into the same category. "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

7. Challenge of the Word-Tree ● In practice however, this approach caused the grouping of a large number of unrelated posts ● Because certain common tags are repeated. ● Solution: Pruning ○ Ranked the tags by their occurrence and deleted 10% of the most commonly used tags. ○ This leads to reasonable separation of post categories. ○ Each tag category was encoded with a positive whole number and applied to the posts. "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

8."Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

9. Methodology 1. Automated Data Scraping 2. Feature Selection 3. Tag Grouping 4. Feature Learning with Stacked Auto-Encoder 5. Classification by Multilayer Perceptron "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

10. Automated Data Scraping ● Data from the GQ India Instagram account was extracted using the API provided by Instagram. ● 32 requests per invocation in a JavaScript Object Notation (JSON) ● 65 features collected for a total of 1280 entries or posts. ○ 1280 X 65: 83200 data-points ● Quantization of data: ○ number of likes in the post were granulized to groups of 25 ■ Eg. likes between 0-25 were labeled as Class 1, and so on. "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

11. Feature Selection Filter Applied Creation Time Week Of The Year Day of The Week Hour of the Day Image (JPG) Caption Length of Caption Number of Tags Tag List "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

12. Why is time of the post needed? "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

13. Auto Encoder ● 4 Layer stacked autoencoder is used to obtain an optimal representation of the data. ● This is extracted from the ‘Z’ layer as “features”. ● Extracted features are classified with a Multi-layer perceptron. "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

14. Results "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.

15. Conclusion Network is able to deliver accuracy of classification higher than 88%. With a granularity of 25 likes per class, this performance is acceptable for commercial applications such as prediction of popularity of a sponsored post, hence price fixation. In the future, this system can be improved using computer vision and CNN based techniques to enrich the input features. "Predicting the popularity of Instagram posts for a lifestyle magazine using deep learning." In Communication Systems, Computing and IT Applications (CSCITA), 2017 2nd International Conference on, pp. 174-177. IEEE, 2017.