WEBVTT

00:00:00.000 --> 00:00:04.680
Hey everyone this is Venkata Pagadala. Today I'm at the Knowledge Management conference

00:00:04.680 --> 00:00:11.120
2026 talking at Taxonomy Boot Camp. So today my subject is Let Customers Name

00:00:11.120 --> 00:00:15.480
Your Categories. So I'm going to explain with the real-time examples and the

00:00:15.480 --> 00:00:21.840
process how to do that and steps to avoid and steps to follow and this is a

00:00:21.840 --> 00:00:28.840
foundational level of organizing knowledge or data. So I'm from

00:00:28.840 --> 00:00:34.600
Atlanta I flew in on November 15, yesterday, which is my birthday as well so I was

00:00:34.600 --> 00:00:40.080
like finding the best cuisine restaurant to have a nice dinner. So I

00:00:40.080 --> 00:00:44.400
found the restaurant which has 20,000 reviews by the way and found it might be

00:00:44.400 --> 00:00:49.240
good and just curious on finding what are the many options they have and

00:00:49.240 --> 00:00:54.120
seeing what are the reviews that actually customers are leaving about and then I

00:00:54.120 --> 00:00:58.480
found that people are mentioning about crab cakes, oysters, calamari which are my

00:00:58.480 --> 00:01:03.280
favorite by the way and that's what I ordered yesterday. So I was curious on

00:01:03.280 --> 00:01:09.160
seeing like what are the actual names that restaurant have and had and then

00:01:09.160 --> 00:01:14.800
what are the customers are referring to and how Google is classifying this. So if

00:01:14.800 --> 00:01:19.480
you can see here the actual names are different but the customer mentions are

00:01:19.480 --> 00:01:24.040
different and the customer mentions are how the Google is classifying them. Even

00:01:24.040 --> 00:01:28.240
the category can be same but you can see the difference between the actual

00:01:28.240 --> 00:01:34.120
mentions and the real names that restaurant had. So that's pretty much

00:01:34.120 --> 00:01:40.440
about all my process of collecting data classifying them and mapping them. How to

00:01:40.440 --> 00:01:47.040
do a knowledge management is the whole subject of this presentation. By the way

00:01:47.040 --> 00:01:52.960
I really love the calamari. So I'm going to drop notes, transcripts, slides,

00:01:52.960 --> 00:01:59.440
everything here so pretty much you guys can follow the video and if you want any

00:01:59.440 --> 00:02:04.800
downloadable, no email required you can directly access this. So your customers

00:02:04.800 --> 00:02:08.000
across the whole industry have already searched for you. What I mean by

00:02:08.000 --> 00:02:14.040
this is like we are not trying to find a new customers or like new search

00:02:14.040 --> 00:02:19.640
patterns. One or other way one or other platform your customers are like and

00:02:19.640 --> 00:02:23.180
future customers or potential customers are already searched for product or

00:02:23.180 --> 00:02:30.440
service. So how can we collect those data? Which are the platforms we

00:02:30.440 --> 00:02:34.400
need to collect the data from and how do we classify them and what are the common

00:02:34.400 --> 00:02:39.800
mistakes we do. So before jumping into all the details and processes I want to

00:02:39.800 --> 00:02:45.240
highlight the importance of the graphical relation of our data. So if we

00:02:45.240 --> 00:02:53.480
see Google, Microsoft, AWS, Amazon, IBM, Palantir, OpenAI pretty much all top

00:02:53.480 --> 00:03:00.440
companies are doing one common thing is a graphical relation. Again this is

00:03:00.440 --> 00:03:10.960
not something net new. If you go to like maybe 20-30 years ago everyone used to

00:03:10.960 --> 00:03:17.560
have a map of address where they need to go. So it's simple I see in a simple way

00:03:17.560 --> 00:03:21.960
you have a bunch of data around and how do you map in order to reach what you

00:03:21.960 --> 00:03:26.880
want is it's a foundational level of explanation of how the graphical

00:03:26.880 --> 00:03:33.160
relation works in data. That's the important that's this tells me the

00:03:33.160 --> 00:03:39.720
importance of graphical relations. So again who I am and what is my

00:03:39.720 --> 00:03:44.400
credibility to talk on this. So let me introduce myself I'm Venkata Pagadala. I

00:03:44.400 --> 00:03:49.840
work in search industry for more than a decade. I worked with brands like

00:03:49.840 --> 00:03:55.960
AT&T, Apartments.com and rehabs.com and researched with [unclear], Previsible,

00:03:55.960 --> 00:04:01.320
[unclear], Amazing Digital, [unclear], and I write research papers and

00:04:01.320 --> 00:04:08.560
find the new patterns of it. That's my credibility to talk for today. So:

00:04:08.560 --> 00:04:13.520
how I started, how I found the sources, from source to product, from tree to graph,

00:04:13.520 --> 00:04:20.280
how I score it, and how to start. So where did it started from

00:04:20.280 --> 00:04:23.800
is like as I work in enterprise companies there's a lot of different

00:04:23.800 --> 00:04:28.120
parts or teams working parallel the one common question they ask like hey do we

00:04:28.120 --> 00:04:32.680
have a page for this? What are the topics we're trying to launch or belong to? Is

00:04:32.680 --> 00:04:36.880
there a demand for that? So why do we have like and more pages on specific

00:04:36.880 --> 00:04:40.760
topic? Is this because with demand or accidentally created by multiple teams

00:04:40.760 --> 00:04:47.000
unknowingly? So these are the common things that commonly received by team

00:04:47.000 --> 00:04:51.920
members across the board and sometimes from the leadership as well. So I decided

00:04:51.920 --> 00:04:57.000
to answer them all with one simple solution. I analyzed the whole industry

00:04:57.000 --> 00:05:05.680
of 5.56 million queries then I kept them into like 1.9

00:05:05.680 --> 00:05:10.840
millions with uncounted questions and then group them into topics of group

00:05:10.840 --> 00:05:18.920
them into topics and then find the top level L1 queries. And if you see this

00:05:18.920 --> 00:05:23.800
like from 5.56 to I cleaned up 3.7 millions of noise that is

00:05:23.800 --> 00:05:30.840
crucial for knowledge management because accuracy as much as we can plays a key

00:05:30.840 --> 00:05:36.360
role because a wrong classification can dilute the whole project. So I'm going

00:05:36.360 --> 00:05:43.120
to explain the steps to avoid as well. So let's take in one example cheapest

00:05:43.120 --> 00:05:48.680
unlimited plan for two lines near me so you can see there are like four kind of

00:05:48.680 --> 00:05:53.640
signals there. Unlimited plan is an entity cheapest is a need compare is

00:05:53.640 --> 00:06:00.240
the intent and modifiers are near me. So how can I get these queries from? So

00:06:00.240 --> 00:06:05.080
again people leave digital footprints all of the place if you see from Google

00:06:05.080 --> 00:06:09.520
search we have a lot of tools directly from first party and third party data

00:06:09.520 --> 00:06:13.760
sources as well and you can see reviews you can see the ratings you can see the

00:06:13.760 --> 00:06:18.080
actual post comments again I'm not going to explain each and everything here

00:06:18.080 --> 00:06:22.760
source by source but you can see search and discovery answer engines video audio

00:06:22.760 --> 00:06:27.000
social and communities like Reddit plays a key role in finding some queries

00:06:27.000 --> 00:06:31.840
helps to classification and experts in long forms like you can find all these

00:06:31.840 --> 00:06:38.000
queries data sets to start with it can be a product or service on a topic so I'm

00:06:38.000 --> 00:06:42.480
going to take an example of Google My Business reviews so I was curious so I

00:06:42.480 --> 00:06:46.080
was curious on analyzing what are the common pain points on telecom industry

00:06:46.080 --> 00:06:50.960
for local stores example if you go for any local telecom brands what are the

00:06:50.960 --> 00:06:55.280
pain points in industry wide not in a specific brand so I was analyzing

00:06:55.280 --> 00:07:00.240
hundred thousand reviews in especially one location Texas and then I found like

00:07:00.240 --> 00:07:04.200
these are the reviews and these are the pain points so what are the pain point

00:07:04.200 --> 00:07:11.560
that commonly mentioned a thousand times which is that second to the most

00:07:11.560 --> 00:07:20.080
pain points is like a long wait so how do they classify this so before jumping

00:07:20.080 --> 00:07:26.320
into classification like all this if I does not have a one view to understand

00:07:26.320 --> 00:07:30.960
the relation between the topics I would really have a hard time to understand

00:07:30.960 --> 00:07:38.720
connect the dots so I dumped everything and created an relational graph then so

00:07:38.720 --> 00:07:43.600
'long wait' is the node, and from the graph I understood that this comes

00:07:43.600 --> 00:07:48.760
under Support & Account and store experience store experience and

00:07:48.760 --> 00:07:53.520
wait time that's how we classify that so the source is finding from reviews and

00:07:53.520 --> 00:07:58.960
the graphical relation is a node and then product is it can be defined by

00:07:58.960 --> 00:08:03.000
their companies but example would be virtual appointments digital check-ins

00:08:03.000 --> 00:08:07.560
virtual assistants are like an agent that can help to resolve the things

00:08:07.560 --> 00:08:13.400
faster before the human talks to them to reduce the wait time that's how if you

00:08:13.400 --> 00:08:17.960
classify the data like find the source classify it and build a product that's

00:08:17.960 --> 00:08:22.720
how the return on investment happens but if you don't classify them in a right

00:08:22.720 --> 00:08:26.800
channel every time the data comes it's stored in different place and cleaning

00:08:26.800 --> 00:08:31.680
them and reorganize reorganizing the entire thing takes as much as time so

00:08:31.680 --> 00:08:35.160
four levels of structure category sub category topic bucket keyword node and

00:08:35.160 --> 00:08:39.880
then how do we say like hey is the four fixed no it depends on the queries

00:08:39.880 --> 00:08:44.600
example would be one query classified what earns a level and what does not so

00:08:44.600 --> 00:08:51.600
if you see here the query is best clear MagSafe case for iPhone 17 Pro Max under

00:08:51.600 --> 00:08:57.120
$30 near me for my daughter this is a common way people search in AI or like

00:08:57.120 --> 00:09:03.000
Google so then I have to go from what is the core product that what is that the

00:09:03.000 --> 00:09:07.080
foundation level would be device hardware and what does it comes under

00:09:07.080 --> 00:09:12.520
accessories, and in accessories they clearly mention about iPhone 17 pro max then I

00:09:12.520 --> 00:09:18.120
have to go for accessories, then cases and protection and then in a phone case

00:09:18.120 --> 00:09:23.520
and [unclear] iPhone case again these can be structured

00:09:23.520 --> 00:09:29.560
in a way possible but keep in mind that it should have a right levels of it you

00:09:29.560 --> 00:09:36.280
cannot have accessories above device and hardware or you cannot have cases and

00:09:36.280 --> 00:09:41.200
protection above the phone [unclear], because the data should flow in a way

00:09:41.200 --> 00:09:48.040
that it can flow from step one to step two but not from step two to step one so the

00:09:48.040 --> 00:09:56.480
process to reach the end goal should be a structured flow so yeah this is how

00:09:56.480 --> 00:10:02.520
it becomes a tree from a tree to a graph once you collect all the data and you

00:10:02.520 --> 00:10:07.040
find the neighbors this is how it becomes so these are the four simple

00:10:07.040 --> 00:10:14.000
rules: known names, web address, meaning, and AI checks. so the label was

00:10:14.000 --> 00:10:17.960
written after reading the word never before so here in example if you see

00:10:17.960 --> 00:10:23.600
like speed test so speed test is in what people typing for and then you need to

00:10:23.600 --> 00:10:28.400
understand what does the speed test belongs to primary category would be

00:10:28.400 --> 00:10:32.120
internet and broadband and again you can go on multiple classifications you can

00:10:32.120 --> 00:10:36.480
say upload rate download rate you can go as specific as possible I'm just

00:10:36.480 --> 00:10:43.520
explaining the foundation level similarly a few other examples what what a rule

00:10:43.520 --> 00:10:47.960
based first pass actually did so we have a rules and in some cases like Nintendo

00:10:47.960 --> 00:10:51.800
Switch is not related to telecom industry but because of a switch it can

00:10:51.800 --> 00:10:57.120
have classified into switching similarly Ring camera and all this other stuff as

00:10:57.120 --> 00:11:03.800
well so even Family Dollar near me can be classified into stores and location

00:11:03.800 --> 00:11:10.520
but that is not the telecom industry of classifications so how like how to score

00:11:10.520 --> 00:11:14.840
that and how to filter it out score equal to 100 minus the percentage of

00:11:14.840 --> 00:11:20.000
with no real home so example we need to find and filter and use a loop to clean

00:11:20.000 --> 00:11:25.280
this up that's that's how you can see the treated topics that's how very the

00:11:25.280 --> 00:11:31.160
more noise you can clear it up the more bucket with generalized buckets you can

00:11:31.160 --> 00:11:37.700
clear it up the better it is again it's gonna be really hard to reach hundred

00:11:37.700 --> 00:11:43.560
so realistically having some decent number and improving every time every

00:11:43.560 --> 00:11:51.200
percentage our loop is expected but I don't expect to be 100 100% so this is a

00:11:51.200 --> 00:11:58.440
stats of the process and these are the details of that and twelve kinds of mistakes,

00:11:58.440 --> 00:12:04.360
each one with a gate: matched one part of a word these are some basic

00:12:04.360 --> 00:12:10.640
steps to follow how can you level set the orders how do you make sure the

00:12:10.640 --> 00:12:16.680
there is no duplicates and how can you avoid having a gigantic buckets one word

00:12:16.680 --> 00:12:21.280
several meanings empty topics so these are the some of the things that to avoid

00:12:21.280 --> 00:12:25.080
so these are the something some of the basic things we can find like no

00:12:25.080 --> 00:12:29.080
duplicate names one version only levels in order levels in order plays a key

00:12:29.080 --> 00:12:35.840
role and this these are the things to follow I did this particular thing on

00:12:35.840 --> 00:12:42.300
education healthcare and also home services I I classified them tested this

00:12:42.300 --> 00:12:49.560
one found a good result so far and I'm still experimenting more on how can be

00:12:49.560 --> 00:12:54.120
structured fast and relatively faster and better so the checklist here is a

00:12:54.120 --> 00:13:00.780
checklist collect, clean, read, name, level, sort, score and check, and repeat so

00:13:00.780 --> 00:13:05.840
that's the whole story of it like taxonomy a tree of categories

00:13:05.840 --> 00:13:11.200
thesaurus, one name for each thing plus the other words; ontology is and relation

00:13:11.200 --> 00:13:19.200
the knowledge graph is relation at scale so this is a couple of sources you can

00:13:19.200 --> 00:13:23.320
read through and I'm personally right now reading Ontology Pipeline, a really good

00:13:23.320 --> 00:13:29.560
book to follow so that's all from me today and thanks for your time
