This document is the master index and combined digest of all 20 lecture PDFs in the Information Retrieval course. Each lecture has its own detailed note (summary_01.md … summary_20.md) containing: (1) a topic table-of-contents + short gist, (2) key terminologies with meanings & use cases, (3) a Mermaid process/workflow diagram, and (4) ELI5 explanations of complex terms.
| # | Summary File | Source PDF | Topic |
|---|---|---|---|
| 01 | summary_01.md | 01_Intro_IR.pdf |
Introduction to Information Retrieval |
| 02 | summary_02.md | 02_UserInt.pdf |
User Interfaces for Search |
| 03 | summary_03.md | 03_Modeling_Part-I.pdf |
IR Modeling I — Boolean & Vector Models, TF-IDF |
| 04 | summary_04.md | 03_Modeling_Part-II.pdf |
IR Modeling II — Probabilistic & Alternative Models |
| 05 | summary_05.md | 04_Evaluation_IR_Part-I.pdf |
Retrieval Evaluation I — Precision, Recall & Metrics |
| 06 | summary_06.md | 04_Evaluation_IR_Part-II.pdf |
Retrieval Evaluation II — DCG, BPREF, TREC |
| 07 | summary_07.md | 05_RF_QE.pdf |
Relevance Feedback & Query Expansion |
| 08 | summary_08.md | 06_Doc-Query.pdf |
Documents & Queries — Languages & Properties |
| 09 | summary_09.md | 07_Text_class.pdf |
Text Classification |
| 10 | summary_10.md | 08_Embedding.pdf |
Embeddings — LSI, Word2Vec, CBOW, Skip-gram |
| 11 | summary_11.md | 08ex_Embedding_LSI.pdf |
LSI — A Worked Example |
| 12 | summary_12.md | 09_MIR_Part-I.pdf |
Multimedia IR I — Image & Audio Retrieval |
| 13 | summary_13.md | 09_MIR_Part-II.pdf |
Multimedia IR II — Audio Features, Video & MPEG-7 |
| 14 | summary_14.md | 10_Indexing_Part-I.pdf |
Indexing I — Inverted Indexes & Query Processing |
| 15 | summary_15.md | 10_Indexing_Part-II.pdf |
Indexing II — Searching, Construction & Structure |
| 16 | summary_16.md | 11_WebIR_Part-I.pdf |
Web Retrieval I — Web Structure & Architecture |
| 17 | summary_17.md | 11_WebIR_Part-II.pdf |
Web Retrieval II — Ranking, Spam & Interaction |
| 18 | summary_18.md | 12_Neural_Ranking.pdf |
Neural Ranking Models |
| 19 | summary_19.md | 13_WebCraw_Part-I.pdf |
Web Crawling I — Concepts & Architecture |
| 20 | summary_20.md | 13_WebCraw_Part-II.pdf |
Web Crawling II — Scheduling Policies & Crawlers |
flowchart TD
A[01 Intro to IR] --> B[02 User Interfaces]
A --> C[03-04 IR Models
Boolean/Vector/Probabilistic]
C --> D[05-06 Evaluation
Precision/Recall/TREC]
C --> E[07 Relevance Feedback
& Query Expansion]
A --> F[08 Documents & Queries
languages & properties]
F --> G[09 Text Classification]
C --> H[10-11 Embeddings & LSI
Word2Vec]
F --> I[12-13 Multimedia IR
image/audio/video]
F --> J[14-15 Indexing
inverted indexes]
J --> K[16-17 Web Retrieval
PageRank/HITS/spam]
H --> L[18 Neural Ranking
BERT/DRMM/DUET]
C --> L
K --> L
K --> M[19-20 Web Crawling
policies & politeness]
M --> K
D --> K
E --> K
Foundations (01–02). IR is finding unstructured material that satisfies an information need from large collections, ranking by relevance. It contrasts with exact-match data retrieval. The user side matters too: search is dynamic (“berry-picking”/orienteering), and the user interface (SERP, snippets, suggestions) shapes perceived quality.
Models & Ranking (03–04). The classic models are Boolean (set logic, no ranking), Vector (TF-IDF + cosine similarity, partial matching), and Probabilistic (rank by probability of relevance, leading to BM25). Alternatives include LSI, language models, and neural networks.
Evaluation (05–06). Effectiveness is measured against reference collections (Cranfield paradigm) using precision/recall, MAP, P@k, MRR, F-measure, graded DCG, and BPREF, scaled up by TREC and the pooling method, complemented by user-based A/B testing and clickthrough analysis.
Improving & Representing (07–11). Relevance feedback (Rocchio) and query expansion (thesauri, clustering, clicks) refine queries. Documents & queries rest on markup languages and statistical laws (Zipf, Heaps, entropy) plus preprocessing (stopwords, stemming). Text classification (Naive Bayes, SVM, kNN) organizes documents. Embeddings (LSI via SVD, Word2Vec CBOW/Skip-gram, BERT) move from sparse to dense semantic representations.
Media, Scale & the Web (12–20). Multimedia IR bridges the semantic gap for image/audio/video (CBIR, fingerprinting, MPEG-7). Indexing with inverted indexes makes retrieval efficient. Web retrieval adds the Web graph, link-based ranking (PageRank, HITS), learning-to-rank, web spam, and massive distributed architectures. Neural ranking (representation/interaction models, BERT) is the modern state of the art. Web crawling feeds the whole pipeline, governed by selection, re-visit, and politeness policies.
| Concept | Appears in | One-line meaning |
|---|---|---|
| Relevance | 01, 05, 06, 18 | Degree a document satisfies the user’s need — the central IR notion |
| TF-IDF | 03, 08, 14 | Term importance = frequency-in-doc × rarity-across-collection |
| Inverted index | 01, 14, 15, 16 | Word → list of documents (the core efficient structure) |
| Precision & Recall | 05, 06, 09, 17 | Relevant-retrieved ÷ retrieved / ÷ all-relevant |
| BM25 | 04, 17, 18 | Modern probabilistic ranking with TF, IDF, length normalization |
| Zipf’s & Heaps’ Laws | 03, 08, 14, 16 | Word-frequency skew / sub-linear vocabulary growth |
| Relevance feedback | 02, 07, 09, 18 | Use marked/clicked results to improve the query |
| Embeddings | 10, 11, 18 | Dense vector representations capturing meaning |
| PageRank / HITS | 04, 17 | Link-based authority/quality ranking |
| Clickthrough data | 06, 07, 17 | Implicit relevance signal from user clicks |
| Semantic gap | 12, 13 | Distance between raw media signal and human meaning |
| Politeness / robots.txt | 19, 20 | Rules so crawlers don’t overload sites |
Generated from the 20 course PDFs. Open any linked summary_NN.md for the full per-lecture note (TOC + gist, terminologies, Mermaid diagram, ELI5).