Facebook Topic Data - FAQs

Facebook topic data is a complex, real-time data source. On this page we cover frequently asked questions that customers have.

Why do I see the the text '[REDACTED]' in my analysis results?

When you analyze links contained in your index you may see results such as the following:


The text '[REDACTED]' has nothing to do with the redaction and quantization rules we apply to analysis results.

Instead this text appears because Facebook removes 9-digit, 10-digit and 16-digit numbers from story text and links to prevent personally identifiable information (such as including US phone numbers, social security numbers and credit card numbers) from being revealed. You may also see this behaviour in the content of posts when analysing super-public text samples.

Why do I see a discrepancy between volumes reported by the pylon/get and pylon/analyze endpoint?

Due to the way data is received from Facebook in to the PYLON platform, you may see small discrepancies between the count of interactions in an index, and the count of interactions in the result of an analysis call.

We receive interactions from Facebook who store the underlying data in a number of partitions. Data from each partition is streamed to the platform separately, and as each stream can see variation in latency, the order of the data we receive is not strictly in the order of events occurring on Facebook.

When you start a recording any interactions recieved by the platform, from the recording start time (and which match your interaction filter) will be recorded to your index. As there may be a small latency on one of the streams of data we receive, a small number of interactions that occurred a short time before the recording start time may be recorded to your index.

This is in contrast to analysis requests, which typically you will specify an exact start and end time for. Analysis will be performed for the exact period you request.

Therefore, as your index could contain interactions from a short time before your the time you started recording, whereas your analysis call has an exact time period, the count of interactions may not match exactly as you might expect.

This emphasises how important it is to specify a time period for every analysis query you make.