> ## Documentation Index
> Fetch the complete documentation index at: https://woku.app/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Text analytics with AI

> Automatic classification of comments, sentiment, keywords and discovery of emerging topics with AI in open-ended responses

**Text analytics with AI** turns your customers' open-ended
comments (text or transcribed voice notes) into structured, actionable
information. It works on the responses from **woku, NPS, CSAT and CES** and
combines three capabilities: **automatic classification** of each comment,
**keywords and summaries** generated by AI, and **discovery of
emerging topics** across the set of responses.

<Note>
  All analysis is automatic and continuous: it is applied to each response at
  the moment it comes in, with no prior configuration and no rules to
  maintain.
</Note>

## Automatic classification of comments

The AI engine classifies each comment, whether written text or a
transcribed voice note, into one of two categories:

* **Recognition**: the customer highlights something that worked well.
* **Improvement**: the customer points out something that can be fixed.

This classification is applied on its own as each response comes in, across
the four tools (woku, NPS, CSAT and CES), and lets you immediately separate
what your customers celebrate from what bothers them, and
quantify the proportion of each one.

To organize responses by your business attributes (branch,
campaign, order, agent), you use **external trackers**, which travel
attached to each response. The
[alerts](/docs/en/configuracion/reglas-de-alerta) notify you by email about the
status of your goals and the response volume, and the escalation of
negative feedback to [customer support tickets](/docs/en/guias/tickets-sac) is
automatic.

## Sentiment

Each response falls into one of three sentiment levels, **critical**,
**neutral** or **positive**, derived from the rating the
customer left. These levels feed the analysis counts: each discovered
topic shows its distribution of critical, neutral and positive responses,
which lets you see at a glance whether a topic concentrates frustration
or praise.

### Our own research

woku combines **market-leading language models** with
its own research in sentiment analysis for Latin American Spanish,
developed in collaboration with the **PhD in AI** of
the University of Concepción (UdeC), the Federico Santa María Technical
University (USM), the University of Bío-Bío (UBB) and the Catholic
University of the Most Holy Conception (UCSC), the first PhD in Artificial
Intelligence in Latin America.

<Card title="Interview about the model's development" icon="linkedin" href="https://www.linkedin.com/posts/pauriquelme_ia-gowoku-ugcPost-7465134849452908545-Pwaa/?utm_source=share&utm_medium=member_desktop&rcm=ACoAAArMdJkBjSyLzdF_dxZqLfZdeR81u-xX-Pg">
  Video with the interview of the PhD in AI academic about the development of woku's own sentiment model.
</Card>

## Keywords and summaries

**Keywords** are extracted automatically from each piece of feedback. These
words feed the **word cloud** in the woku, folder and NPS reports, which
also include an **AI-generated summary** and **suggested actions** based on
what customers say.

CSAT and CES have their own **AI Summary**, which synthesizes in natural
language the responses of each measurement.

## Emerging topics

Beyond the categories you already know, woku's **semantic engine**
groups responses by meaning and discovers **emerging topics** that
were not defined in advance, with **exact** counts: each figure
corresponds to real responses you can review.

Topics are explored in the **reviews space**, a 3D universe where
each response is a point and responses that talk about the same thing end
up close together. Selecting a topic opens a panel with:

* **Label and summary** of the topic.
* **Number of responses** that make it up.
* **Sentiment distribution** (critical, neutral, positive).
* **Trend** and first and last appearance over time.
* **Evidence**: the real responses that support the topic.

The analysis is unified: a single topic can bring together responses from woku,
NPS, CSAT and CES.

## Where it is displayed

Text analytics results appear in several places across the
platform:

* **woku, folder and NPS reports**: word cloud, AI summary and
  suggested actions.
* **Report builder**: the feedback type (recognition or
  improvement) is available as a dimension to cross it with other
  metrics. See the [report builder](/docs/en/reportes/builder-visual).
* **Reviews space**: visual exploration of the emerging topics and their
  sentiment distribution.
* **Data Studio**: qualitative questions, such as what my
  customers are talking about, are also answered with reports whose numerical
  evidence is verified against the data. See [Data Studio](/docs/en/reportes/data-studio).