HYRUM-S/PAI

this is my first machine learning project leveraging machine learning , natural language processing , digital image processong as i implement them in a cross-site automated sentiment analyser shopping assistant. currently focused on the Jumia shopping website one of the leading africa e-commerce website # Project PAI: Cross-Site Shopping Assistant & Smart Advisor An autonomous, API-independent, multi-disciplinary e-commerce smart intelligence proxy engineered entirely within an optimized **Arch Linux** host environment. The system resolves consumer informational imbalances—such as review manipulation, astroturfing, and artificial price distortions—by acting as an automated full-stack computational engine. It handles high-dimensional product spaces entirely without official, vendor-filtered APIs, translating noisy web layers into a single, manipulation-resistant utility score ($R$). --- ## Theoretical Grounding & Problem Space ### 1. Bounded Rationality & The Curse of Dimensionality Evaluating a single product tier across multiple online storefronts (e.g., Jumia, Amazon, AliExpress) requires human shoppers to process thousands of unstructured data vectors—ranging from volatile base pricing and logistical variables to qualitative, potentially fraudulent review text arrays. While classical economic models presume an unbounded rational agent executing deductive logic to maximize utility ($\max_{x \in X} U(x)$), human cognitive capacity is strictly bounded by metabolic and attention limits (~15 minutes mean manual calculation time). This choice proliferation scales the space exponentially to $\mathcal{O}(M \cdot N^K)$ ($M$ platforms, $N$ products, $K$ attributes) , overwhelming human focus boundaries and driving severe conversion deficits and cart abandonment. ### 2. The Consumer Contextual Decision-Making Model (CCDMM) Project PAI anchors its workflow within Suomala's CCDMM framework. Human cognition utilizes two core behavioral paths when interacting with online marketplaces: * **Similarity-Strategy (SIMS):** A low-energy default where incoming marketplace inputs match prior internal models ($C_m$), allowing rapid, low-energy decisions based on stable heuristics. * **What-is-Out-there-in-the-World-Strategy (WOWS):** Triggered when anomalous data spikes (cont …