AAI_2025_Capstone_Chronicles_Combined

Cinema Analytics and Prediction System

23

Figure 21: LSTM + GloVe model similarity score

The LSTM model’s recommendations combine linguistic embeddings and numeric

metadata, capturing genre, theme, and contextual patterns. Figure 22 shows PCA-reduced

features from the model’s penultimate layer, visualized with five K -Means clusters, matching

the cluster count used for TF-IDF and BERT, to compare semantic groupings across models.

Figure 22: LSTM-based movie clusters

The resulting clusters show varying degrees of thematic coherence (Figure 23):

190

Internal