Hi there 🖐️ , here there are some of the projects I did work for:
For this project I have been asked to collect Twitter data using twarc in order to find out which was the most popular language in the dataset and also to calculate the average, minimum and maximum number of followers that users in the dataset have.
The most popular language in the dataset was English and the average number of followers of the users was 29.5K, the minumum number of followers was 0 and the max was 8.5 million.💪
Data Exploration | Data Cleaning
This was another challenge from the Twitter project 1. I have been asked to answer the following question: to what extent does the sentiment expressed in a tweet influence user engagement with the tweet (likes and retweets)?
A Sentiment Analysis has been performed (on Azure) on the Tweets. The dataset extracted has been merged witht the original dataset and consequently analysed: as a result, the average number of times that a Tweet has been retweetted or liked was higher when the sentiment expressed in a Tweet was Positive rather than Negative or Neutral sentiment. The positive sentiment expressed in a Tweets positively influence user engagment.👌
Data Cleaning | Merging Data | Sentiment Analysis
As a conclusion of the Twetter venture, I have been asked to perform some Automated Content Analysis on the Twetter data in order to analyse and estimate how many of the users were possibly willing to participate to a fundrising campaign organised by the ONG at the base of this project.
The results of the Automated Content Analysis showed that 10 tweets out of 110 were referring to the 50th anniversary of the company, about 9.1%. Instead, 9 tweets out of 110 were referring to a fundrising or supportive-related topic, about 8.1%. The data of the two variables have been merged in a third variable, namely "possible_targets", showing that out of a sample of 110 tweets 17 are referred to enthusiastic users who were mentioning the ONG anniversary or either topics which can be in favour of a fundrising campaign, about the 15.5%. The results show that a possible 15.5% of users among the social medis followers of ONG could be targeted by the fundrising campaign since they would be presumably willing to support the campaign, although, the sample is too small to accurately predict the final result. Considering the whole amount of social media followers of the ONG (31 millions), if a 15.5% would approximately contribute to the fundrising campaign with 1 dollar,an estimation of the result of the Fundraising Campaign can be deducted to 4.805.000,00 $. 👍 👀
Automated Content Analysis | Categorization | Tokenization | Data Visualization
For this project I have been asked to work with a dataset from an online store that created a set of marketing campaigns. The data related to the purchases made throughout the campaigns will be used to perform predictive analysis and create a model to forecasts whether a specific campaign or referral better stimulate users to make a purchase or not (purchase) and, if so, how much they would be willing to spend (order_euros).
The data collected showed that among 8479 observation, 2239 users did make a purchase, only the 26% of them. The three main referrals are Google, Instagram, Facebook and all the other observations have been categorised as "other" since the listed platforms were not providing a significant amount. Facebook showed that 28% (each percentage here it's about the platform's specific users and thus are not cumulative between different platforms) of the users have made a purchase and the average amount spent was 54€. Google presented a 57% purchase rate and the average amount spent was 109€. Instagram presented a 6% purchase rate and the average amount spent was 10€. Regarding the advertising campaigns, the first campaign (campaign1) showed a 53% purchase rate and the average amount spent was 106€. The second campaign (campaign2) showed a 17% purchase rate and the average amount spent was 29€. The third campaign (campaign3) showed a 16% purchase rate and the average amount spent was 28€. The fourth campaign (campaign4) showed a 17% purchase rate and the average amount spent was 28€. In conclusion, the most profitable referral was Google, either in terms of purchase rate and amount of euros spent through the orders. Instagram was the less profitable referral, either in terms of purchase rate and amount of euros spent. On the other hand, the most profitable marketing campaign resulted being Campaign 1. 😎
Classification | Binary variables | Data Modeling (Linear + Logistic Regression) | Predictive Analytics
For this project I have been asked to work with a set of different marketing campaigns and perfrom data analysis in order to understand if an endorser would better enhance users to make a purchase and, if so, how much they would be willing to spend (order_euros). For this challange, I will focus on predictive analytics and data modeling. Moreover I will be evaluating the performances of models and perform some machine learning in order to make the model more accurate.
The marketing campaign with no endorsers (CPC campaign) seem to be the best among all the others. The predictions made shown that the chances of making a purchase are 51.5% if a user has seen the CPC campaign. Instead, if a user has seen the influencerA marketing campaign, the chances of making a purchase are 28.2%. Also, if a user has seen the influencerB marketing campaign the chances of making a purchase are 32%. Additionally, the amount of the purchase in € looks better stimulated by the CPC campaign as well: 95€ is the average purchase made from a user who interacted with CPC vs 60€ from the influencerB campaign and 28€ from influencerA campaign.👌
Predictive Analytics | Data Modeling | Machine Learning | Performance Metrics | Decision Forests
For this porject I have been asked to answer the following question: to what exextent a fundraising campaign with endorsers will better enhance donations and consequently increase revenue compared to a fundraising campaign without endorsers?
The coefficients of model B show that, compared to other campaigns, celebrity (- 0.99) and influencer's (- 1.00) campaigns significantly and negatively influence donations and they do no significantly differ from each other. Instead, a campaign with no endorser (coefficient 2.30) shows that, compared to other campaigns, it significantly and positively influence donations. The model proved that, compared to other campaigns, influencer's campaign did not result better than celebrity or no endorser campaigns. Instead, a campaign with no endorsers resulted to be the most successful. Predictions from model B show the likelihood of making a donation among different campaigns: with all the control variables kept constant, the campaign with no endorsers has been proven the most successful on stimulating users' donation with a 75% chance of making one. Results showed that, compared to other campaigns, celebrity (-6.16€ per each donation) and influencer (-6.26€ per each donation) campaigns negatively and significantly enhance the likelihood of donating an higher amount of euros. A campaign with no endorsers (+34€ per each donation) positively and significantly enhance the likelihood of donating an higher amount of euros. 🤓💫
Data Exploration | Data Cleaning | Merging Data | Data Visualization | Data Modeling (Linear + Logistic Regression) | Predictive Analytics | Machine Learning | Performance Metrics | Decision Forests