Let's begin with the caveats. Much of filter performance, like search performance, is dependent on transcription quality. Machine transcription is fairly impressive for what it is, but it is what it is. Users with better transcriptions will have better experiences.
Then the technicals. Groups of filters are between horizontal lines. Within groups, filters operate as ORs - the more filters you select, the more hits you will have. Across groups, they operate as ANDs
- Filters can be used in combination with or independent of text search terms.
- If used in combination, we suggest you run your text search first, and then drill down using filters.
- All textual analysis is done at sentence, rather than whole video, granularity.
When moving from one search query to the next, please use the Clear Search button next to the search bar.
Owner, Source, Type
These filters allow you to restrict your results by item type - for example just the videos you have clipped out of the originals AND placed in a playlist.
Who's in it
As we don't have per-respondent metadata for your sample, we use facial recognition technology to capture gender, age and number of respondents, along with any "observed emotion" that can be captured through analysing facial responses.
What's being said
The filter groups under this set use text analytics to infer the content and context of the video, so are of course dependent on transcription quality. They are sub-grouped into what we hope are logical sets. Please note that if you run a text search for "chicken soup" and set a filter for "taste", the highlighted hit will be on the soup, and the taste - which may be salty or delicious or similar - will refer to the video item in its entirety. The operator between multiple filter sets in What's being said is an AND; within sets, it's an OR.
Stated feelings, unlike "observed emotion" above, are derived from textual analysis. Noteworthy among these filters is our sentiment analysis - positive, negative, and neutral - where any particularly strong sentiment expressed by the respondent is identified.