At the end of my EE writing process, certain considerations came to my mind. The first possible improvement was the sample size of data used, both for Neural Network training and comparing the NN with the algorithm. Using larger data sets might’ve reduced the chances of error lower than I’d predicted, but like all scientific experiments 100% accuracy is never achievable. Secondly, the problem of overtraining the NN as it reacts differently to different types of data sets, showing that the NN might find unintended sarcasm, this is inversely relatable to the algorithm’s ability to detect irony. However, my results answered my research question and gave insight into my original goal to efficiently settle any political debate with the help of computers. My research can help companies wanting to efficiently mine customer trends and have ethical consequences like election vote swinging. My research begged the question “What is privacy today with the widespread use of data mining to mould our every choice?”. Computer science must have ethical considerations, and this has opened a new area of interest for me, I want to conduct further personal research on this and even study it in university.
EE
Interim Reflection (R2)
People’s varying political views urged me to see if a computer could provide an accurate analysis. My previous algorithm was unsuitable due to comparison issues, so I used IntenCheck API and Keras/TensorFlow neural network, which are robust pieces of software that output data in a comparable format, on a scale of 0 to 100 for the value of every sentiment in text. I feel that the strongly opinionated tweets were a good way to test the accuracy of both pieces of software. Since analysing sentiments is always subjective, I used consensus to minimise the errors in classifying the tweets’ data sets. Then checking their accuracy by comparing results, I took the ratio of their results to draw up extendable conclusions. My decision to switch the algorithm and overall focused research approach will benefit me in overcoming challenges in future research papers as well. Conducting this experiment has increased my knowledge of AI, such as limitations in power consumption and storage.
Reflection 1 (R1)
After careful considerations and preliminary research, I formulated my research question. Being from an Indian background, whenever I talked to people, it always troubled me that every next person had a conflicting opinion on political parties. So, to find the most efficient way to analyse these opinions, I decided to research on Sentiment analysis, I’m comparing the neural network approach to the algorithmic approach, using Tweets. So far the biggest obstacle I encountered was choosing a robust piece of software for each approach. Which I overcame after pre-testing and solving installation bugs. I’m looking forward to conducting my experiment because it widens my knowledge in the field of A.I. while also benefiting me on a personal level. For me, the next hurdle would be neural network training which I am looking forward to. I believe my research question is also strong due to the amount of benchmarking data available.