Ask HN: Do you know of any good product taking advantage of NLP?

76 points by julienreszka ↗ HN
Natural language processing seems very popular today but is it actually any good and useful? When was your last encounter with a good product taking advantage of this technology?

64 comments

[ 2.9 ms ] story [ 115 ms ] thread
Interested as well, but I keep feeling nlp is relatively less stable and mature compared to vision, thus maybe only used in niche ways rather than full fledged products, other than obvious ones like echo and siri of course
Almost any google search you do or any siri question you ask takes advantage of the NLP research that has been produced in academia over the last decades.

Theoretical research of these subjects it's not like an open-source project used in industry: "who's using this tool to do x, y and z?". The output of NLP research is the combination of thousands of various papers that, all combined, makes something like a siri search possible.

Is Siri any good ? I don't use siri I don't know, do you think it is?
Siri isn't very strong, it can answer (very) basic questions and resorts to web searches when it can't answer. I find it occasionally mishearing me or completely parsing my question incorrectly. I recall a particularly bad one: "set a timer for 3 minutes and 15 seconds" and somehow it interpreted it as an arithmetic question "the answer is 18."

Alexa and particularly Google Assistant are much better, the latter retaining context between questions e.g: "who is the 44th president of the US" followed by "who is his wife?" is a neat trick.

No way. Hello Google is miles better. Google got the upper hand on AI currently. You can see it with their camera AI technology Pixel 3 is top dog in camera even though it have one len.

But Apple just took 2 famous Google AI researcher. So it seems that Apple is trying to catch up.

I believe Apple's issue is lack of data.
Oh really? Okay, I just read how two Google's top AI researchers went over to Apple so I just assume their problem was lack of AI expertise. But I can see your point since Apple isn't a software company that hoard massive amount of data like Google is doing.
facebook, they are scanning your chats trying to figure out what is your political party, social status and needs so they can sell you more stuff, more expensive and more often.
right. but is it actually any good at it?
They are amazing at it, just look at the salaries they pay and the shock and denial when you tell the hoi polloi what they are up to.
A lot of (the better) customer service/support chat bots use NLP to assist users.
Can you please give the names of the products that you are talking about?
There are a decent number of companies that use NLP voice recognition when you call their support lines. So when you call, instead of pressing a number on your key pad to move through the directory you can speak it.

In that same vein, I've seen companies that have "chat support" use NLP bots to get a feel for the type of question a person is looking for before handing off to a human support person.

I've also seen AI personal assistants that handle meeting schedules etc like https://x.ai/

Translation is a NLP problem... Isn't it? I've used Google translate to read non-english websites and found it quite useful. Is text to speech a NLP problem? Today's TTS products are actually listenable these days.
Gmail.

It gives suggestion on reply message. I use it a lot.

NLP can be use in informational retrieval such as organizing things into cluster of topics via LDA.

I've met someone with the mindset that AI is a fad and that what have it done so far? I disagree, it may be over hype but I see that person using Siri. I use Hello Google and I consider that to be NLP not the traditional statistic NLP but the deep learning NLP kind.

Also those website translator such as google translator and old school babel translate helped me a lot when I was trying to search for things that were in Japanese. There were a few famous Korean and Chinese novels that are machine translated and it's decent enough to read if you're really into that novel and willing to ignore the quality.

Interesting to hear that - I actually find Gmail's autoreply feature to be extremely annoying. When it's not suggesting outright inappropriate responses (once had it suggest "Ok! see you there!" on a thread about a funeral a colleague was attending), they just seem trite and useless.
I also don't like this feature. I've used it only perhaps a handful of times since it was introduced, but every other time it has actually cost me the time it took to read and mentally reject the suggestion.

I really need to poke around the settings and see if it can be disabled...

They recently updated it and it is much more relevant in my case. (More business than personal)

Keep in mind features like this are built around the needs of the people who make it. Outlook is sort of a snapshot of the needs of a Microsoft employee circa 1996. GMail is a little better, but is still of reflection of the company.

Text Analytics such as: Topic Modeling (extract the topics from customer reviews / interviews / text documents)

Sentiment Analysis (get the tonality from the entire text or the tonality from specific parts of the text)

Disclaimer I work at a web application / API text analytics company: https://gavagai.io/

Here are a Quora post my CTO speaks more about our tool and analyzing survey text data.

https://www.quora.com/Whats-an-automatic-efficient-way-to-an...

https://www.quora.com/What-is-the-best-approach-to-quantify-...

Neural machine translation (NLP with deep neural networks) has become quite good and is increasingly put into useful products.

(Disclaimer: I work on DeepL translator)

Very good work with DeepL. I'm often in contact with professional translators and they are very impressed.
Does DeepL work better than Google Translate? For Spanish and Russian?
I'm building a Chrome extension[1] which uses NLP to determine whether you're on task by comparing the topics contained in any new page opened.

It works fairly well, albeit not perfectly at this point.

[1] - https://deepmode.app

Looks really cool extension. I would like to give it a try. I tried to click the Subscribe button but it is not working.
Damn that's annoying, my apologies. Will fix as soon as I can get to my laptop.
I'm the founder at Akia (https://akia.ai). We're using NLP to help hotels with the number of inbound they receive from guests who are coming on or staying on property.

Though I'm not personally a regular user, from what I've observed, the AI is able to very, very efficiently resolve a significant number of requests because a lot of the messaging is the same: "what's the wifi?" "bring me towels" "can i get my car?" etc.

The application of NLP for a niche industry has felt particularly appropriate for a small startup where in contrast, doing generic NLP really feels like a boil the ocean type strategy that makes more sense to those who have the resources (google/apple).

In the second screenshot, there are suggested messages. One of them says "Good afternoon, Mr. Washington...". Does your app understand gender?
Custom Voice Commands is an iPhone and Android smartphone app that uses NLP to allow you to tag photos, videos, notes and web pages with questions or phrases. For example, you can tag a bunch of photos with the phrase "show me my favorite lunch photos" and every time you speak that phrase you'll see a slideshow of those photos.

Photo Recall is an Amazon Alexa and Google Assistant app that lets you retrieve Unsplash photos by voice plus photo slideshows created with Custom Voice Commands can be viewed from your Amazon Echo Show or Google Home Hub (or via the Google Assistant app).

https://treycent.com/app/content/photo-recall.html

Text classification is widely used in almost any major website, for example for spam detection, putting things into categories, detecting inappropriate content, ...
Seems like it's too obvious as it's a chat bot, but I think meetcleo is pretty good (disclaimer: I work there)
It depends on what you mean useful. My startup tracks everyone’s mood, activity, and experise on HN. Checkout yourself:

https://hnprofile.com/author_profiles?utf8=&search=julienres...

This can be used to great effect in companies. Essentially, we deploy a search engine for tribal knowledge in a company. Overall, we see significant improvements company wide. Reduced turnover and in turn improved knowledge retention

That's amazing ! Could help many people who are lonely to better know about themselves
Whoa.

So have you already started analyzing word patterns and finding different accounts people have on the same service and also correlating them between multiple services?

People do leave much more information about themselves on the internet than they realize.

(comment deleted)
Past NLP researcher chiming in here. Google Home and Alexa use pretty sophisticated speech to text software to decouple your voice from background noise, and then translate our speech to text, and then from text to program triggers that ie turn off your lights. It may sound trivial but the Home work so well it really is kind of magical.

The Chatbot stuff is more questionable, especially the ones based on deep neural nets ( again I've published in this domain so I looked at it pretty closely ).

Twitter/FB/Douyin/Google all use NLP to tune their feed or give you search results.

Just to give you some examples...

I suppose spell checking (and then grammar too) has been the most ubiquitous use of NLP to date. Unless I'm missing something even more pervasive (software).
https://www.chowly.com/

They are a great service to restaurants and use NLP to automate a lot of service integration such as between the embedded POS system and delivery companies.

NLP is used in a lot of marketing tools. Check out crimson hexagon, infegy atlas, netbase
Most search engine products, be it job search, Airbnb, e-commerce, Google, etc are going to have a lot of opportunities and use for NLP to understand queries and content.
Besides the obvious applications like search and voice assistants here's a few really good ones

* https://www.grammarly.com/ is pretty interesting. Corrects your grammar and not just your syntax

* Swype keyboard which uses your spelling corrections as a way to better figure out how you use a swype keyboard

* https://flexibits.com/fantastical and many other calendar apps do NLP stuff to pull out meeting info. Microsoft, Google and Facebook do this kind of stuff pretty well (Disclosure: I worked on such features at Microsoft)

Happy to give more examples if people are interested in working in this space.