section_link_analysis.md

Link analysis

Notes:

• How to use link analysis to determine relevant pages?

---

Assumption

• In-link = endorsement
• Similar to citation analysis

Notes:

---

Link info

• Anchor text describes target page
  • "computer manufacturer" → lenovo.com
• Also text surrounding links
  • "Find funny cat pics here"
• May also be abused
  • "evil empire"

Notes:

---

PageRank

• Rank pages with many in-links higher
• PageRank =~ In-degree
• But harder to influence as it propagates
• One in-link from high PageRank site worth more than many low PageRank in-link
• Probability that random surfer will end up on page

---

PageRank vs In-Degree

---

Pagerank calculation

1. Assume random surfer
2. Randomly walk web graph
   • No out-links → teleport
   • 15% chance that user opens a random page
3. Count how many times a page is visited

$$\sum \text{pagerank} = 1$$

---

Pagerank visualization

Notes:

---

Reasonable surfer

• Clicks some links more often than others
• Main content vs sidebar / footer
• Anchor text related to query / user intent
• Avoid ads

Notes:

• How can the random surfer be improved to provide more realistic results?