A novel model to cluster emojis

I'm looking for a novel model to create pseudo labels for a tweeter dataset. Some may use hashtags, some may use emojis. Some may combine few approaches.(it's called distant supervision as well). The purpose is to eliminate manual labeling, and yet create a reliable labeled dataset, therefore it should improve the existing works and have better results. In my project I want to use emojis ( you may add other features, but emoji is necessary to be a part of it) So I want to know what is the emotion of each emoji, cluster emojis into 11(eight Plutchik + optimism, pessimism,love) classes and assign an emotion class to it (joy, ecstasy, sadness,...) Later I may extend 11 classes to 28 emotion classes The proposed mode/method is better to be based on word embedding approach (/ word2vec) and it must improve the results of the benchmark. Similar works to get an idea (for your own read) : Distant Supervision for Emotion Classification Task using emoji2emotion Beyond Just Text: Semantic Emoji Similarity Modeling to Support Expressive Communication Learning Emoji Embeddings using Emoji Co-occurrence Network Graph [login to view URL] [login to view URL] **Benchmark: Any of the above, you can choose any other recent benchmark, we'll discuss about it What I'm expecting is a code(Python) and documentation. Documentation includes : - A table of similar approaches (10 studies), with limitation of each approach. it can be around 1000 words. - A complete explanation of the "new" model, it's advantages over existing approaches, and justification of its "uniqueness" (at least 1500 words) You have to explain the flow clearly, you may use figures. - Evaluation of the proposed model and justify the improvement compare to the "benchmark"( I gave code and paper of the benchmark) The code must include: -Comments (very important) and I need a clear representation of structure of code. Duration: 09 days. Budget: $230 Tool: Python OS: Windows 10