Practical Recommender Systems For Business Applications

Explore the pandas library to structure data with series and data frames, learn how to read data from external sources into data frames, and perform basic data wrangling and analysis.

Read multiple csv files from a single folder, concatenate them into one dataframe, and create a new column to identify each listing’s origin, then clean the combined data.

Read in data from an SQLite database using Python: connect to the database, create a cursor, explore tables, and execute select statements to retrieve and print table names and rows.

Learn to read JSON files and understand the JavaScript object notation, including area, location, description, image, and hectare criteria, and read data from local and web sources.

Learn to read text data in Colab, including unstructured data, by mounting drive and using open-based methods. Cover two quick file-reading approaches in Google Colab.

Assess data quality by detecting missing values with Caleb, dropping empty columns, and exploring positive and negative correlations to guide imputation, cleaning, and prep for predictive modeling.

Learn to pre-process data with Python and pandas by dropping columns, handling nulls, and imputing ages and fares with means for modeling.

Group data by category using group by to summarize by country and subdivision, compute mean, count, and other descriptive statistics, and generate aggregated views of university rankings and rainfall data.

Master data summarization and pivoting with movie metadata to uncover insights. Pre-process columns, handle missing budgets, and compute mean revenues by production countries and by collections using pivot tables.

Explore basic data visualizations on the 100k movie dataset to understand ratings, age distribution, gender distribution, and movie genres, guiding practical recommender system insights.

Explore more data visualizations in a recommender systems context, using Yelp reviews to map star distributions and city level ratings with interactive plots.

Examine the temporal dimension of movie metadata by analyzing monthly releases and revenue trends, using date parsing, month order, and plotting.

Explore principal component analysis as an unsupervised dimensionality reduction method that converts correlated predictors into uncorrelated principal components, capturing maximum variation and enabling further analysis.

Apply principal component analysis in Python by scaling numerical iris dataset features, compute four components, and use the explained variance to reduce predictors for regression or classification.

Learn the theory of singular value decomposition (SVD), decomposing a matrix into three matrices, and apply SVD for feature extraction, decomposition, and image reconstruction in recommender systems.

Apply singular value decomposition to extract two components for dimensionality reduction, compute singular values, and visualize the reduced iris dataset in two and three dimensions.

Explore unsupervised learning theory through imagery analysis in remote sensing, where spectral bands cluster data without labels to reveal land use patterns like elevated ground and industrial areas.

Explore unsupervised clustering with k-means to partition data into clusters around centroids, and learn the iterative assignment of observations to the nearest centroid by Euclidean distance.

Explore cosine similarity and cosine distance between vectors using Python, demonstrating a simple computation with two example vectors and the role of these measures in recommender systems.

Learn to compute the Jaccard score between two vectors, use macro average, and derive the Jaccard distance as 1 minus the score for practical recommender systems.

Learn how supervised classification uses training sites and spectral signatures to classify remote sensing pixels, using algorithms like minimum distance, maximum likelihood, and spectral angle mapper.

Apply k-nearest neighbours using euclidean distance to classify by majority of neighbors. Explore support vector machines with kernels and ensemble methods like random forests and gradient boosting to enhance performance.

Explore content-based filtering and collaborative filtering, including memory-based and model-based approaches. Learn hybrid filtering, item- and user-based methods, clustering, neural networks, spatial networks, and association techniques with Python implementations.

Compute euclidean distance between users' ratings to identify the shortest distance that reflects similar preferences and use these distances as the basis for recommendations.

Learn how to combine distances and singular value decomposition (SVD) to build movie recommendations, constructing and normalizing a ratings matrix, applying cosine similarity, and identifying top similar movies.

Showcases how demographic filtering uses demographic data, similarity based on age, gender, and nationality, and neighborhood selection to tailor recommendations.

Merge movie datasets by id, rename and select key fields, then compute mean votes and apply an 85th percentile cutoff to generate recommendation candidates.

Apply IMDb scoring to filter top movies for recommendations using a weighted rating and percentile filtering, and sort by score to surface the top titles.

Explore a crude correlational approach to item-based filtering by merging ratings and books data, building a user-item matrix, and identifying items correlated with a given book to drive recommendations.

Install and use the surprise package to build book recommendation models, load ratings data, apply cross-validation with multiple algorithms, and compare similarity indices for better content filtering.

Explore a hybrid recommender that blends content and collaborative filtering with the LightFM package. Encode users and songs, build a co-occurrence matrix, and generate user-specific song recommendations.

Set up a basic pre-processing scheme to prepare JSON datasets for classical content filtering and collaborative filtering, extracting attribute dummies from restaurant attributes in Mesa.

Apply content-based filtering by using item attributes to build a simple nearest-neighbor model, evaluate it with train-test splits, and identify restaurants most similar to Taco Bell.

Explore collaborative filtering with a subset of reviews to build a user-item rating matrix, then use a correlation matrix to identify five restaurants similar to red and white blue.

The lecture explains how to clean text data using the beam package, applying lower, remove punctuation, strip whitespace, remove numbers, remove stop words, and stemming to build a document matrix.

Convert the tweet text to a single string, lowercase it, and remove usernames and links to prepare data for a word cloud in a step-by-step text cleaning workflow.

Lowercase the text, remove Twitter usernames, and keep only alphabetic words for analysis. Remove words shorter than three characters and strip extra spaces to enable word cloud generation.

Develop NTLK-based cleaning skills to improve data quality for practical recommender systems used in business applications.

Apply a nltk-based tweet data cleaning workflow by defining a text cleaning function, normalizing case, removing punctuation and stop words, and performing stemming and lemmatization for downstream analysis.

Discover word clouds as a visual, frequency-based representation of textual data, highlighting topic keywords. Learn to build word clouds from text using generators or Python.

Visualize movie themes with word clouds built from titles and overviews. Build a corpus by joining titles and overviews, remove stop words, and highlight the most common terms.

Learn tf-idf theory by computing term frequency and inverse document frequency with a log-based weighting, and apply it to textual data in natural language processing.

Apply tf-idf to extract insights from text, transform sentences into vectors, and compile a data frame of feature names and idf values, illustrating its use in recommender engines.

Apply content-based filtering on text data with the surprise package, cleaning titles, building tf-idf features, and computing cosine similarities to surface similar book recommendations.

Word2vec uses neural networks to learn word associations from text, detecting synonyms and relatedness to power item-based recommendations from product descriptions.

Explore another approach to finding similar items and outline a more detailed recommender system using the same dataset, pre-processing steps, and a similarity model to suggest items like white variants.

Build a word2vec based item recommender by processing purchases, building a product vocabulary, training a model with negative sampling, and recommending similar products via vector similarity.

Explore dictionaries as Python's mutable key-value data structure, learn to declare, access, add, delete, copy pairs, and use len, in, not in, and sorted.