leksara 0.1.3

Library pemrosesan teks Bahasa Indonesia untuk domain e-commerce (cleaning, PII masking, review mining, pipeline).

Leksara

Leksara is an Indonesian-language text preparation toolkit for data teams who need production-ready cleaning, masking, and normalization pipelines. The library bundles linguistic resources, a preset-driven orchestration layer, and modular helpers so you can audit raw text, remediate sensitive content, and standardize noisy reviews without rebuilding the stack for every project.

---

Feature Highlights

• CartBoard review intake – Inspect raw datasets from chatbots or marketplaces, generate column-level flags (PII, non-alphabetical noise, ratings), and capture metadata for monitoring.
• Composable cleaning utilities – leksara.function re-exports the building blocks (HTML stripping, casing, stopwords, punctuation, emoji, numeric cleanup) for ad-hoc preprocessing.
• PII masking and redaction – Regex-backed detectors for Indonesian phone numbers, emails, addresses, and national IDs with configurable replacement modes and conflict handling.
• Review-focused normalization – Slang and acronym expansion, contraction repair, elongated text trimming, rating extraction, stemming/normalization tuned for Bahasa Indonesia.
• ReviewChain orchestrator – Run pipelines functionally with leksara(...) or via the ReviewChain class, mix presets with custom steps, and benchmark per-stage performance.
• Resource-driven customization – Ship your own dictionaries and regex rules or extend the bundled JSON/CSV files to adapt the cleaner to new verticals.

Deep dives for each module live in docs/features.md together with API tables, dependencies, and ready-to-run recipes.

---

Quickstart

1. Install

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install leksara

Optional extras and troubleshooting tips are listed in docs/installation.md.

2. Clean a review column in-place

import pandas as pd
from leksara import leksara

df = pd.DataFrame(
    {
        "review_id": [101, 102],
        "review_text": [
            "<p>Barangnya mantul!!! Email saya [e-mail], WA 0812-3456-7890</p>",
            "Kualitasnya ⭐⭐⭐⭐, pengiriman 4/5. Hubungi [phone]",
        ],
    }
)

# Apply the ecommerce review preset
df["clean_text"] = leksara(df["review_text"], preset="ecommerce_review")
print(df[["review_id", "clean_text"]])

3. Audit raw text with CartBoard

from leksara.frames.cartboard import get_flags, get_stats

flags = get_flags(df, text_column="review_text")
stats = get_stats(df, text_column="review_text")

print(flags[["review_id", "pii_flag", "rating_flag", "non_alphabetical_flag"]])
print(stats.iloc[0]["stats"])  # nested histogram of noise sources

4. Compose a tailored pipeline

from leksara import ReviewChain
from leksara.function import (
    case_normal,
    remove_punctuation,
    remove_stopwords,
    replace_email,
    replace_phone,
)

chain = ReviewChain.from_steps(
    patterns=[(replace_phone, {"mode": "replace"}), (replace_email, {"mode": "replace"})],
    functions=[case_normal, remove_stopwords, remove_punctuation],
)

cleaned, metrics = chain.transform(df["review_text"], benchmark=True)

---

Documentation Map

Topic	When to read	Location
Installation & environment	You are provisioning a workstation or CI agent	docs/installation.md
Feature deep dives	You need behavioral details, configuration knobs, or per-feature dependencies	docs/features.md
Public API reference	You want signatures, argument descriptions, and return payload formats	docs/api.md
Worked examples	You prefer copy/paste recipes for notebooks or pipelines	docs/examples.md
Dependency matrix	You must vet optional packages or align with enterprise policies	docs/dependencies.md
Contributing	You plan to submit patches, run tests, or build docs	docs/contributing.md

---

How Leksara Fits Together

• Pipelines – leksara(...) is a convenience wrapper around ReviewChain; both accept raw sequences (list/Series) and return cleaned text plus optional benchmarking details.
• Frames layer – CartBoard and friends operate on review tables, deriving flags, statistics, and noise diagnostics suitable for dashboards.
• Functions layer – The leksara.function module mirrors the implementation modules under leksara/functions so you can cherry-pick individual cleaners without touching internals.
• Resources – Regex rules and dictionaries stored under leksara/resources/ drive PII detection, slang resolution, and whitelist protection. Update these files to specialise the toolkit.
• Logging & benchmarking – leksara.core.logging ships opt-in helpers to emit step-level logs, while benchmark=True collects timing metadata for throughput tuning.

Architectural notes, data contracts, and extension points for each layer are captured in docs/features.md.

---

Contributing & Support

• Read docs/contributing.md before opening a pull request. It covers environment setup, style, testing, and documentation requirements.
• File issues on GitHub with reproducible examples; include the preset, optional dependencies, and OS details when reporting pipeline differences.
• Commercial or large-scale users should build automated smoke tests around ReviewChain to detect upstream dictionary or regex changes.

Leksara is licensed under the terms specified in LICENSE.