Normally, when we talk about computer vision, artificial intelligence, or deep...learning, we're discussing hard philosophical problems (strong AI). PornHub, however, has its own uses for the technology: It wants to use it to improve video tagging, including accurately identifying performers and offering a more intelligent set of criteria than the typical tags you find on video.
This is a rather sticky problem that doesn't just impact those who keep lube manufacturers and online stripper shoe stores in business. If you've been online for any length of time, you've probably encountered a similar issue at least once or twice. While Google is an excellent search engine overall, there are times when the terms that describe what you're searching for are too generic to return useful results. At that point, teasing the appropriate data set out of the search engine's tight grip becomes more art than science.
Google's occasional problems are just the tip of the iceberg. Porn sites often have extensive category-based libraries of content, but the sheer size of their databases would overwhelm even the most anal librarian. Newegg--yes, Newegg--actually has a great example of how categorization can fail to deliver the results customers were begging for.
Newegg has an option called "Power Search" that allows the end-user to choose specific traits from a wide range of options, in the hopes of drilling down into results and nailing the proper product. Most of the time it works quite well, but take a look at the company's "Compatibility" section for its heatsink and fan hardware.
Understanding why the above is such a problem requires some knowledge of heatsink design, so I'll explain. Intel and AMD periodically introduce new CPU socket designs with different physical dimensions. Both companies are actually pretty good at maintaining backwards compatibility with older heatsinks. [Only you, Joel, can somehow find a way to talk about Intel, AMD, CPU sockets, and heatsinks in a story about porn. -Ed] While not every Athlon 64 cooler was compatible with AMD's later motherboards, heatsinks that relied on a latching mechanism back when Socket 754 and Socket 940 were new (2003) can still use those coolers ten years later on Socket AM3. Similarly, Intel has kept its desktop sockets mechanically identical for eight years.
The problem with Newegg's database is that it's based on whatever the manufacturer put on the packaging, even if there's absolutely no reason why a heatsink that fits a certain CPU wouldn't fit others in the same family. The boxes in the image above show three examples of different heatsink categories that are functionally and practically identical. Any heatsink that supports LGA1155 supports LGA1156. Any heatsink that supports Socket AM2 also supports Socket AM3. Any heatsink that supports Socket 939 also supports Socket 940. And any CPU that supports Socket 754 supports Socket 939.
In a situation like this, a search feature that's meant to offer end-users fine-grained control over their searches ends up doing the opposite. Clearly a different approach is needed--ideally an AI that could swallow whatever enormous database of porn it gets fed while spitting back accurate results.More Fap, Less Crap
PornHub supports tagging and categorization, but the one-armed handed bandits that make up the site's clientele tend to either leave tags off or stuff generic labels on to every video. PornHub's new AI has been trained to recognize the faces or other body parts of professional performers. End users will also be allowed to give input on the accuracy of the new system, upvoting or downvoting the AI's accuracy. The site only intends to label performers who work in the space professionally; amateurs will not be tagged with their names by the new AI.
"Now, users can search for a specific porn star they have an affinity for and we will be able to retrieve more precise results," PornHub Vice President Corey Price said in a statement.
Our sister site PCMag(Opens in a new window) also notes PornHub hopes to scan all 5 million videos in its database, and is planning to expand the AI's ability to identify other aspects of a scene. The company is working on stretching the AI's capabilities to include organization by position, location, and hair color as well.
"Artificial intelligence has quickly reached a fever pitch, with many companies incorporating its capabilities to considerably expedite antiquated processes," Price said. "And that's exactly what we're doing with the introduction of our AI model."Tagged In
Pornography Tagging Pornhub Categorization AiMore from Internet & Security
