Reply to: Have you done Natural Language Understanding?

[DimPrawn]

This is the most applicable API I have found so far for text analysis

MonkeyLearn - Machine Learning for Text Analysis

[Cirrus]

Maybe you need something like Watson: https://personality-insights-livedemo.mybluemix.net/

I had a look at MS LUIS as suggested above but it looks more like pulling out themes from lots of documents rather than 'recognising' the type of individual document. So it would be good at finding out what themes are connected with Adele's left foot on social media or saying "If you're interested in Adele then everybody is talking about her left foot."

Watson seems a bit similar. If you asked it "how many feet does Adele have when she's holding a baby?" it might get the right answer. However the challenge I have is : here are 10000 emails and the category they have been assigned to - can you spot the pattern?" I've only seen that with Neural Networks eg card fraud detection.

[original PM]

Someone in India will read them and scoop all the personal data all for $10 per hour.

FYFY

[DimPrawn]

Your central mailbox is the one I'm talking about. 60,000 emails per month - all requiring back office FTEs to read and route the mail to an appropriate queue. It costs a lot of money.

Someone in India will read them all for $10 per hour.

[darmstadt]

Maybe you need something like Watson: https://personality-insights-livedemo.mybluemix.net/

[Alias]

Your central mailbox is the one I'm talking about. 60,000 emails per month - all requiring back office FTEs to read and route the mail to an appropriate queue. It costs a lot of money.

What you need is Natural Language Processing. ServiceNow can do that up to a point but you really need a specialist to be able to parse the emails on the fly and decode the "language" and infer meaning before deciding what it needs to do with it. There are a few companies out there that specialise in it (not cheap!). BBC used one for the Olympics as a way of using semantics with annotation in publishing.

[Cirrus]

Now if you need to route the initial email from a customer, that isn't coming in from the contact us on the website (which should have a reason dropdown), then you should have all initial contacts going to a central mailbox, where the code above is inserted by a person

Your central mailbox is the one I'm talking about. 60,000 emails per month - all requiring back office FTEs to read and route the mail to an appropriate queue. It costs a lot of money.

[DimPrawn]

Exactly. The person in the client company that is responding to the email from a customer must know what the business purpose of the email is (renewal, cancellation, amend, etc). When they reply that can embed a code or keyword into the response body or subject line,e.g.

Regards,

Dave


###-467-###

where 467 means something to the routing and the ###- and -### is a marker for the code as an example.

You just need to work out the business process.

Now if you need to route the initial email from a customer, that isn't coming in from the contact us on the website (which should have a reason dropdown), then you should have all initial contacts going to a central mailbox, where the code above is inserted by a person, and then the system does the routing appropriately on any further customer replies.

Either that or you have separate email addresses for customers to make the first contact:

enquiries@clientco.com

renewals@clientco.com

cancellations@clientco.com

etc, and hope customers use the correct email most of the time

[original PM]

I have been looking for something like this since about 2005 and until true AI is available I do not think it can be done accurately.

So the answers as have already been cited in this thread is a combination of

1) Get customers to use the contact us page of your website will assist especially if you give them a list of 'reasons' for the contact. There is some rationale behind as people will more likely remember your web address or be able to find it in google than your e-mail address

2) Obviously if you have some form of online customer portal where people log in and can submit tickets this will also enable effective routing.

3) If they are responding to an e-mail sent by you ensure your subject line has some form of way of routing it back to the correct department (again if the e-mail is sent from the mailbox of the department that will deal with the response then no need for routing?)

4) You could write some program which would scan the e-mail and try and route it but the key to making that work is ensuring that when an e-mail gets routed incorrectly there is a clear way to resolve it. All to often I have seen these wrongly routed e-mails end up disappearing down cracks because of an in ability in teams to collaborate. Also the potential cost of managing that ends up overshadowing the savings expected from the routing method.


As for exec's demanding something which may not even logically be possible I have found it is because they have never really had anyone say no to them and so they can just make demands and shout at people until they roll off onto their next 3 year directorship having achieved nothing - still nice work if you can get it.

Or as I like to put it

'It's not fookin magic, Potter'

[Hobosapien]

This is one example where AI will take over.

You'll know you've got a good AI system when it outsources the job to an asian sweatshop after identifying it will be cheaper than the cost of its own electricity per hour.

[DimPrawn]

BI, certainly over most of the last 20 years has been on structured data. I can see you can do that.

We're talking here unstructured.

I really worry that Londonmank used SQL Server to analyse an email. Some disconnect there.

SELECT * 
FROM Email
WHERE MessageBody LIKE '%Wibble%'

Finds all the emails about Wibble. God this is easy.

[Cirrus]

Easy Peasy

BI, certainly over most of the last 20 years has been on structured data. I can see you can do that.

We're talking here unstructured.

I really worry that Londonmank used SQL Server to analyse an email. Some disconnect there.

[Einstein Jnr]

Have you done Natural Language Understanding?

Sounds like you have landed a never ending project - hope you are on a good rate, keep iterating, keep agile and keep billing - I keep hearing we are on the verge of a great break through in AI

[MrMarkyMark]

Business Intelligence is used for Datawarehouse and to create midleware events and alerts. I don't think it will work on this case, if yes will be limited.

[Bee]

All I was saying is that it can very much depend how much you want to achieve with it. Having worked in business intelligence (sometimes an oxymoron) for 20 years, I've had to do a fair bit of data mining. Generally there have been better requirements than "what would you do with this email without know what we'd want to do with it and/or what our company looks like".

Business Intelligence is used for Datawarehouse and to create midleware events and alerts. I don't think it will work on this case, if yes will be limited.