Home / Recipes / Watch what is said about a named individual
Worked queryWatch what is said about a named individual
/stories in hybrid mode. The parameters matter less
than the reasoning behind them, so that is most of what is below.
The request
GET /stories
?q="<full name>" <disambiguating term>
&mode=hybrid
&from=<now - 7d>Why these parameters
Personal names collide constantly, so add a disambiguating term — the company, the role, the sector. Hybrid mode matters more here than for companies because names get transliterated differently in every script and abbreviated differently in every register.
The obvious approach, and why it loses
The obvious approach is the bare name in quotes, and it works until the name is common. Track a chief executive called Chen or Smith that way and you get coverage of everyone who shares it — and because every result is a real article about a real person, nothing in the output looks wrong. The opposite instinct, adding the employer as a second quoted term, overcorrects: a large share of coverage names the individual without naming the company in a form that matches, so the constraint removes the articles you most wanted. A descriptive disambiguator rather than an exact one — the sector, the role, the market — sits between the two and biases ranking without hard-filtering.
When it returns too much, or too little
Too much means the disambiguator is too weak. Try the role rather than the company, since coverage that names a person usually explains who they are, while the employer may appear only in a later paragraph or not at all. Too little is more often a window problem: individuals are covered in bursts around results, appointments and controversies, and a rolling 24-hour window mostly lands in the gaps between them. Widen to seven days before weakening the name itself. Where the name is transliterated from another script, expect hybrid mode to carry most of the recall — keyword matching alone will miss the majority of non-Latin coverage without reporting any error.
How you would know it is working
The honest test here is precision rather than recall, because the two failures are not symmetrical: a briefing note about the wrong person costs more than a missed mention. Sample twenty results and count how many are about the individual you meant. Below roughly nine in ten, the disambiguation is not working and no downstream filtering will rescue it. Keep a written record of why the monitor exists and what is done with its output, separately from the code. Persistent monitoring of a named individual is the one recipe on this site where the compliance question is more likely to end the project than the technical one, and that answer should exist before the pipeline does. Keep the sample recent as well as regular: precision on a name drifts when the person changes role or another public figure with the same name becomes newsworthy, and neither event produces any signal in the pipeline itself.
When this is the wrong tool
Monitoring coverage of a private individual raises obligations that monitoring a company does not. Confine this to public figures acting in a public capacity, and check what your jurisdiction requires before building anything persistent.
Running it for real
Response fields, failure modes, adaptations and scheduling cost: Watch what is said about a named individual, in the docs.