Ambiguous Search Queries: Serving the Right Meaning When 'Apple' Could Be Fruit or Company
Company: Microsoft
Role: Product Manager
Category: Machine Learning
Difficulty: medium
Interview Round: Other
On a general web search engine, many queries can mean more than one thing. Someone who types "apple" could be looking for the fruit or for the technology company. As a product manager, how would you make sure that what each person sees is relevant, meaning it matches what they actually wanted?
```hint Look past the query text
The word alone cannot settle it. Think about what other evidence could shift the odds between the two meanings, and what the page should do when that evidence is weak.
```
### Clarifying Questions
- Is this a general web search engine used by everyone, or a vertical such as shopping or recipes where the context already narrows the meaning?
- What do we know about the person: whether they are signed in, their past searches, their location and device, the previous query in this session? Which of these may we use under our privacy rules?
- Should the answer focus on ambiguous queries specifically, with "apple" as the example, or on relevance for every query?
- Who judges relevance today: user behavior, trained human raters, or both?
### What a Strong Answer Covers
- A definition of relevance that separates which meaning the person intended from how good each page is for that meaning.
- The signals that tip an ambiguous query one way or the other: what most people click for it, the current session, personal history, location and time, and breaking news.
- How ranking and page layout use those signals, including what a good first page looks like when the meaning stays unclear.
- How relevance is measured offline and in live experiments, and why click data alone can mislead.
- Risks such as over-personalization, privacy, and new meanings that appear suddenly.
### Follow-up Questions
- The company announces a major product and searches for "apple" spike overnight. How should results adapt, and how quickly?
- Fruit results get few clicks. Does that prove they are irrelevant, and how else would you check?
- Someone who mostly searches for recipes types "apple stock." How much weight should their history get here?
Overview: A technical product question about keeping web search results relevant when a query such as apple could mean the fruit or the technology company. It tests knowledge of intent signals, ranking and result diversification, the limits of personalization, and how relevance is measured offline and in live experiments.