section_spelling_correction.md

Spelling correction

Palystation, Xboy, Wiii

Notes:

---

Spelling correction

1. Find alternatives
2. Evaluate alternatives
3. Return alternative results

Notes:

• How can we find alternatives?

---

1. Find alternatives

bock~

(Levenshtein, N-Gram)

↓

book, rock, spock

Notes:

---

2. Evaluate alternatives

Notes:

• How can we evaluate alternatives? Which metrics could be applied?

---

Context

• ­ Find alternatives for every misspelled query term → collations
• ­ Search for collation and evaluate results

---

• ­ seach endine?
• ­ seach → peach, search
• ­ endine → ending, engine

---

• ­ seach endine?
• ­ → peach engine? 0 results
• ­ → peach ending? 0 results
• ­→ search ending ? 5 results
• ­→ search engine ? 10 results

Notes:

---

Evaluate alternatives

• Most results
• Best results
• Most searched

Notes:

---

3. Return alternative results

• Feedback
• Transparency

Notes:

• How does this affect precision and recall?
• What kind of spelling corrections does Google offer? What are the typical scenarios?

---

Did you mean

• Original query has decent results
• There is a better alternative

Notes:

---

Instead

• Original query has poor results
• There is a better alternative

Notes:

---

Showing results for

• Original query has no results
• There is an alternative

Notes:

---

Show results with

• No good results including all search terms
• Selectively remove some of the search terms