Muhammad Furquan
Legal & Compliance Reviewer
Muhammad Furquan is a qualified Barrister and legal professional with an LLM from BPP University Law School. He reviews content against copyright law, defamation standards, consumer protection rules, digital publishing guidelines, and broader compliance requirements before publication.
Articles
What Is Information Retrieval? The Core Problem Every Search Engine Solves
Before search engines existed, IR researchers were solving the same core problem: how do you retrieve a relevant document from a large collection? This article defines the field's core concepts, precision, recall, relevance, and the recall-precision tradeoff, grounding every later topic in a rigorous framework rather than marketing folklore.
What Is the Vector Space Model? How Documents Become Numbers (and Why That Changes Everything)
The Vector Space Model represents documents and queries as mathematical vectors, making it possible to compare meaning through distance, angle, and weighted terms instead of simple keyword presence.
TF-IDF and BM25: The Mathematics of Keyword Relevance (And Why Repetition Stops Helping)
TF-IDF rewards terms that appear often in a document but rarely across the collection. BM25 (Best Match 25) extends this with diminishing returns on term frequency and document-length normalisation. Both remain the baseline every modern ranking model is measured against — and understanding them explains why keyword stuffing has never worked.
PageRank: How Brin and Page Replaced Word-Counting with Link-Counting
In 1998, Brin and Page made the leap from word-counting to link-counting. PageRank models a "random surfer" who clicks links with probability d (the damping factor, ~0.85) and occasionally jumps to a random page — the probability of ending up on any page is its rank score. A link from a high-PageRank page passes more authority than one from a low-PageRank page. This lesson covers the formula, convergence, and why this changed the web.
Hubs and Authorities: How Kleinberg’s HITS Algorithm Explains Why Niche Links Beat Generic Ones
Published the same year as PageRank, HITS computes two scores per page iteratively: an authority score (pages pointed to by many good hubs) and a hub score (pages that point to many good authorities). The eigenvector update converges to a stable ranking. HITS explains why topical link clusters matter — and why a link from a domain authority in your niche outweighs a generic high-PR link.
Crawl, Index, Rank: The Search Engine Pipeline That Decides Whether Your Page Exists to Google
Google officially describes three stages: crawling (URL discovery and page fetching), indexing (analysis and storage), and serving (ranking and result delivery). This lesson treats the pipeline as an engineering system with inputs, processes, queues, and failure modes — not just a list of stages. Understanding the whole system before studying each part prevents the tunnel-vision that most SEO courses suffer from.
From Strings to Things: How Google’s Knowledge Graph and Hummingbird Update Changed What “Relevant” Means
The 2012 Knowledge Graph and 2013 Hummingbird update marked the transition from keyword matching to entity understanding. Google now models people, places, organisations, and concepts as nodes in a graph — a query about "Einstein" retrieves the entity, not the string. This lesson explains what entity-based search means for content strategy: topic authority replaces keyword density.
Learning-to-Rank: How Machine Learning Replaced the 200-Factor Checklist
Modern search engines don't hard-code ranking rules — they train machine-learning models on query-document pairs. The learning-to-rank (LTR) field divides into three approaches: pointwise (score each document independently), pairwise (learn which of two documents is better), and listwise (optimise the entire ranked list). RankNet (2005) was the first major neural pairwise model. This lesson introduces the framework that modules 4.4 and 4.5 build on.
MAP, MRR, and NDCG: The Metrics That Define What “Better Rankings” Actually Mean
Before you can improve a ranking system you need to measure it. Mean Average Precision (MAP) averages precision at every recall level. Mean Reciprocal Rank (MRR) measures how high the first correct result appears. Normalized Discounted Cumulative Gain (NDCG) accounts for graded relevance — a result in position 1 is worth more than position 5. These metrics drive every A/B test at Google and every LambdaRank training objective.
The Ethics of Search, the Business Model That Funds It, and What SEO Actually Is
Brin and Page wrote in 1998 that ad-funded search engines have incentives misaligned with user quality. Google's guidelines explicitly separate organic ranking (algorithmic, unpaid) from ads. This lesson covers the ethical framework of SEO — quality, user experience, long-term trust — and debunks the most persistent myths before they take root. It also sets up Google Search Console as the student's ground-truth monitoring tool.
