Best of July

I found the blog "The morning paper", very good. There was a series of papers about chatbot development. The blog is changing topic often by reviewing papers from actual conferences but there are several interesting ones even if it's not about chatbots. Anyway here is the selection about "chatbot" papers:

• On chatbots
• A survey of available corpora for building data-driven dialogue systems
• A neural conversation model
• Incorporating (a) copying mechanism in sequence to sequence learning
• Building end-to-end dialogue systems using generative hierarchical neural network models
• Natural language understanding (almost) from scratch
• End-to-end learning of semantic role labeling using recurrent neural networks
• Machine learning for dialog state tracking: a review
• Multi-domain dialog state tracking using recurrent neural networks
• Ten challenges in highly-interactive dialog systems

Chatbots are not good enough (yet) to replace humans so why do not automate human workforce. Scale (http://www.scaleapi.com/) is a startup who want's to do this. Let get done short tasks by human workers through API calls.

And finally some marketing related infographics from Kissmetrics I came by during the month: How to calculate customer lifetime value:

A more comprehensive list of conversational datasets

The list just the copy of the datasets listed in A survey of available corpora for building data-driven dialogue systems (Serban et al. 2015). All credits goes for them. I simply try to make it more accessible. I couldn't find the download location for every one of them, so if you see a '?' as URL and you know how to get the data, please write a comment about it.

(same google doc as a link)

Best in June

As last month I posted a video about "hyper reality" I also found a writing about "hyper personalization". For me this is just personalization but either way we call it I agree that tere is a move in this direction:

Focus on 3 important ‘W” of hyper-personalization:

• Who you personalize
• What you personalize and
• Where you personalize.

 Anyway, let's switch back to my favorite topic these days, chatbots :) I found a page, it's not exactly about chatbots, but it shows how interesting this simple chat interface can be. There is a weekend project to post interviews as chat one-to-one chat dialog. Check out Talk turkey!

And finally I found a pretty comprehensive list of dialog datasets. I found it in this article: A survey of available corpora for building data-driven dialogue systems (Serban et al. 2015). If you don't want to read the whole 46 pages (btw it's probably worth for you if you are interested in developing chatbots) I'm still processing the list but I will make it available in a separate post soon.

STRATA + Hadoop world London was organized in jun. If somebody missed the keynote from Stuart Russel, I recommend to watch it. He is talking about The future of (artificial) intelligence.

#FUN: And you can also check http://www.projectmurphy.net/ Microsoft's AI bot:

With the Bot Framework you can add the bot on Skype, Messenger, telegram etc.. and ask it all life's most important questions such as: "What if Charlie Chaplin was a baby?" or "What if Beethoven was a rockstar!" Project Murphy then uses artificial intelligence to answer these questions by combining the subject's face with the object of interest i.e. a baby's face smartly added on top of Charlie Chaplin's face.

Best in May

In may there was just a couple of stories grabbing may attention. Here are those:

• Google deputs Allo an AI based chat app. I couldn't try it yet, but it sounds interesting.
• Customer service over chat becoming more and more popular. You don't have to constantly hold the phone and you can talk with others around you while you are chatting with the customer service agent. And of course this is a good business opportunity for chatbots. 

And here is a movie about augmented reality. As I'm working with personalization in retail space, it was particularly interesting for me.

Best in April

Still chatbots and more. Sundar Pichai said this last week on Alphabet’s earnings call:

In the long run, I think we will evolve in computing from a mobile-first to an AI-first world.

Reinforcement learning toolkit for Python: OpenAI Gym

Speech KITT makes it easy to add a GUI to sites using Speech Recognition. Speech KITT provides a graphical interface for the user to start or stop Speech Recognition and see its current status. It can also help guide the user on how to interact with your site using their voice, providing instructions and sample commands.
Rod Humble had previously spent three years as the CEO of Linden Lab, and before that worked on the Sims franchise as an EVP at Electronic Arts. In June, an automated conversation company called PullString (formerly ToyTalk) hired him to create a new series of games for Facebook Messenger called Humani. Here is a nice article about this game: Humani: Jessie's Story.

Chatbot craze - deep learning for chatbots intro

I'm just participating the Deep Learning Udacity course (I really like it :) When I found an article: Deep Learning for Chatbots, Part 1 – Introduction. Exact the topic I'm really interested in.

It's a nice introduction. And it sheds some lights on the capabilities of the state-of-the-art systems. E.g.: "However, we’re still at the early stages of building generative models that work reasonably well. Production systems are more likely to be retrieval-based for now."

Also I found interesting references about incorporating context into generativ models: "Experiments in Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models and Attention with Intention for a Neural Network Conversation Model both go into that direction."

Question Answering datasets

To extend the list of conversational datasets there is a collection of Question Answering (QA) datasets. A question-answer pair is a very short conversation which can be also  used to train chatbots. If you want to use the chatbot for giving information for customers, like automated customer support or automated sales agent on your website, this type of datasets can be particularly useful.

The WikiQA corpus is a new publicly available set of question and sentence pairs, collected and annotated for research on open-domain question answering.

Usually on TREC (Text REtrieval Conference) there is a QA task which has some kind of datasets associated with it. Most of the datasets are focusing on factoid QA task but the one in 2015 is a kind of live QA. The task was to answer questions on Yahoo Answers.

Manually-generated factoid question/answer pairs with difficulty ratings from Wikipedia articles. Dataset includes articles, questions, and answers.

There are some manually curated QA datasets from Yahoo Answers from Yahoo.

You also can download the Stack Overflow questions and answers. It's a domain specific but huge dataset.