Tagging of Words in a Phrase using Word Embeddings; Superblocks in Barcelona: Creating a Pedestrian-Friendly Environment

The technical report details the usage of modern machine learning techniques to make a model more efficient. Work was carried out through a summer internship with Envestnet, from June 2021 to August 2021. Natural language processing techniques were used to optimize a model that used an older, slower, and inefficient method to tag words in a phrase. Envestnet wanted to look into natural language processing to see if any benefits in training speed/efficiency were possible. Natural language processing is an up-and-coming field in machine learning that involves natural language inputs or outputs, such as words, phrases, and sentences. The specific natural language processing technique that I used was called Word Embeddings, which translates words into numerical vectors that carry information about the word and its surrounding context words. The data was organized into credit card transaction phrases, with each phrase containing identifying information about the transaction. Each word could be tagged with six possible options – vendor name, city, state, payment type, store id, and other. By the end of training, words in the phrase were tagged with 95% accuracy using the new model. This accuracy was similar to the original model, but with massive improvements in training time (24 hours vs. multiple days). Next steps could include feeding more data to the new model to increase accuracy, and to predict more tags (currently, 6 tags were used).

The STS research paper explores the implementation of Superblocks in Barcelona. Superblocks are made up of multiple city blocks with vehicles restricted or prohibited on the internal streets – any vehicular traffic is rerouted outside the perimeter of the superblock. This creates a quieter, shared space between city blocks that can be used for a variety of purposes such as gatherings or play areas for children, from space previously reserved for cars. The primary motivation for this paper was to explore how the development of new technologies, such as Superblocks, are influenced by human behavior and social groups. The STS framework used to analyze this was Social Construction of Technology (SCOT), which asserts that human action shapes technology more than technology influences human action. In Barcelona, the Superblocks were a technology that was created by BCNEcologia, an NGO, but evolved to take on different sizes and shapes based on community groups and the rules they decide to enforce.

School of Engineering and Applied Science

Bachelor of Science in Computer Science

Technical Advisor: Daniel Graham

STS Advisor: Travis Elliott