Intelligent information retrieval based on simple implicit learning

{!ctf}

Intelligent information retrieval based on simple implicit learning

Dr Gavin Ruddygav@pontneo.com

17 Feb 2016

Imagine we have some content in a database.

{!ctf}

q1

A user runs a query …

{!ctf}

which finds some results …

q1

{!ctf}

and returns them, sorted in some way.

q1

{!ctf}

The user chooses some of them ...

q1

{!ctf}

which adds to our existing click data.

{!ctf}

When we use this data to help sort results ...

q1

{!ctf}

users can learn from each other ...

q1

{!ctf}

and returns can converge on a consensus.

{!ctf}

q1

We can take this simple idea a step further.

{!ctf}

A user’s clicks happen in a sequence ...

q1

{!ctf}

connecting things intelligently as they make choices.

q1 q2

{!ctf}

Integrating over queries, sessions, users, time etc ...

{!ctf}

maps out the connections users are moving along.

{!ctf}

So when we run our query ...

q1

{!ctf}

we can identify dynamically connected secondary items ...

q1

{!ctf}

and use them to modify or extend our returns …

q1

{!ctf}

or to make intelligent recommendations ...

q1

{!ctf}

or to put ads in the right place at the right time etc.

q1

{!ctf}

These returns get a click response ...

q1

{!ctf}

which adds to our existing click data ...

{!ctf}

so the next time our query runs ...

q1

{!ctf}

new things have been learned.

q1

{!ctf}

We don’t need to actually map all this complexity …

{!ctf}

just query relevant click data at query time ...

{!ctf}

q1

but the effect is intelligent query returns …

{!ctf}

q1

that share the right things in the right places …

{!ctf}

q1

and learn continuously via an implicit feedback loop.

{!ctf}

q1

{!ctf}

No need for machine learning, complex processing or masses of metadata.

To see how this behaves on real data, take a look athttp://www.slideshare.net/pontneo/quick-introduction-to-the-clickthrough-filter