ANLY482 AY2017-18T2 Group30 Instagram

From Analytics Practicum
Jump to navigation Jump to search
APex Logo.PNG


HOME ABOUT US PROJECT OVERVIEW EDA BUSINESS OBJECTIVES PROJECT MANAGEMENT DOCUMENTATION MAIN PAGE
Facebook Post Facebook Video Youtube Instagram Blog Post


Data Source

To retrieve data from the company's instagram, we made use of a web-scraping script from Github. We made modifications to the script to include timestamp as well as caption, the data includes:

  • Caption
  • Timestamp
  • Img URL
  • Tags
  • No. of Likes
  • No. of Comments

Data Preparation

After scraping the data, we realised that the data needed cleaning. The indexes of the column values were off as seen here:

Ig bad data.png

We also concatenated the "tags" into a single column.


Exploratory Data Analysis

The dataset contains Instagram posts collected from August 1, 2016 to February 1, 2018. From the chart below, we can observe that this is a constant decline of number of posts over the time.

Ig posts over time.png

However, despite the decline in number of posts, there is a steady increase of average number of “Likes” over time. This can be possibly due to better quality of posts as opposed to quantity.

Ig average likes over time.png

Analysing the number of posts by Day of the Week tells us that most content are posted on Thursday. There are a considerably less number of posts on the weekends -- which they can consider scheduling their posts on to garner more likes and reach out to more people. We can see that the average number of Likes are spread out quite evenly over the days, which means that more posts should be equally spread out over the days as well. However, the average number seems to be significantly higher on Thursdays.

Ig posts vs likes weekday.png

Further analyses shows that there is an outlier single post garnering over 80,000 likes which has skewed the average. As such we will be removing this data point for further analyses. Upon further investigation, we realised that this is not the number of “likes” for an image, but rather “views”. As Instagram recently added videos to post content, the data will contain videos as well, which cannot be distinguished from images using the current dataset. As such, we will have to make modifications to the crawler script to identify videos and add a metric for “views”.

Ig likes distribution.png

Similarly, we observe the number of posts by the Hour of the Day. Surprisingly, between 12pm to 12am, there are very few posts as compared to the other half of the day. With this, we would like to study if the time of posting affects the total number of likes and comments received.

Ig post over hour.png

Looking at the Average number of Likes over the Hour of the day, there are not much significant differences. The slightly lower number of average likes from 12pm to 12am could be due to insufficient data points.

Ig average likes hourly.png

By plotting a histogram for the number of comments, we see a right skewed distribution. We see that most posts garner around 0-2 comments, with posts rarely having more than 7 comments. Using this data, we created an additional column to categorize the data: 0-2: Low Comments, 3-6: Medium Comments, 7 and Above: High Comments

Ig comments distribution.png

In addition we would also like to investigate if the number of tags in a posts would affect the performance. From the text columns of hashtags, we created a numeric variable which contains the number of tags a post has. The chart below shows a distribution of the number of tags, with most posts having tag counts of 0 - 3.

Ig tags distribution.png