sabin1234/Lemonn_Financial_FAQ

💹 Lemonn Financial FAQ (Nepali) — Dataset README

A 300-pair Nepali-language financial education FAQ dataset — explanatory, paragraph-length answers about mutual funds, IPOs, SIPs, and personal investing, translated/localized into Nepali from source content by Lemonn.

---

🔖 TL;DR (At a Glance)

What	Answer
Total records	300
File size	~657 KB
Language / script	Nepali (ne / npi) — Devanagari
Format	JSONL, conversations (ShareGPT-style, 2-turn)
Task type	instruction-following
License	unknown ⚠️
Data source	Lemonn (an Indian investing/broking platform)
Generation type	real (not synthetic)
Domain	Finance → Stock market basics → Mutual funds, IPOs & personal investing
Market focus	India-specific (despite the Nepali language) — zero Nepal/NEPSE content
Answer style	Long-form, explanatory paragraphs (not short facts)

---

🗂️ Schema — What's Inside Each Record

{
  "id": "lemonn_001",
  "conversations": [
    {"from": "human", "value": "Is mutual fund income taxed?"},
    {"from": "gpt", "value": "Yes, returns from mutual funds in India are taxed. The tax treatment depends on the type of fund ... and the Finance Act 2024 changed certain capital-gains tax rates effective FY 2024–25 ..."}
  ],
  "source": "Lemonn",
  "source_name": "lemonn_all_faq",
  "source_repo": "Lemonn",
  "source_config": "default",
  "source_split": "train",
  "source_revision": "v1",
  "source_row_id": "lemonn_000001",
  "language": "ne", "language_code": "npi", "script": "Deva",
  "license": "unknown", "license_tier": "unknown",
  "task_type": "instruction-following",
  "generation_type": "real", "condition": "real",
  "url": "",
  "metadata_json": "{...}"
}

Field-by-field status

Field	Type	Variation	Value
id	string	✅ unique, sequential	lemonn_001 → lemonn_300
conversations	list[2]	✅ content differs, structure fixed	always human → gpt
source / source_repo	string	❌ constant	"Lemonn"
source_name	string	❌ constant	"lemonn_all_faq"
source_config / source_split / source_revision	string	❌ constant	default / train / v1
source_row_id	string	✅ sequential	lemonn_000001 → lemonn_000300
language / language_code / script	string	❌ constant	ne / npi / Deva
license / license_tier	string	❌ constant	unknown / unknown ⚠️
task_type	string	❌ constant	instruction-following
generation_type / condition	string	❌ constant	real
url	string	❌ empty	always ""
metadata_json	string (nested JSON)	⚠️ one sub-field varies	see below

Inside metadata_json (nested) — one field varies

{
  "generation_domain": "Finance",
  "generation_category": "Stock market basics",
  "generation_sub_domain": "Mutual funds, IPOs, and personal investing",
  "behavior": "Providing informational answers",
  "behavior_definition": "Answering by preserving the original question's meaning and the original answer's facts, rendered into pure Nepali.",
  "question_type": "Informational question" | "Procedural question",
  "content_language": "Nepali",
  "content_script": "Devanagari",
  "english_content_allowed": false
}

Everything here is constant across all 300 records except question_type, which splits:

question_type	Count	%
Informational question (जानकारीमूलक प्रश्न)	223	74.3%
Procedural question (प्रक्रियामूलक प्रश्न)	77	25.7%

The behavior_definition itself is a strong signal about how this dataset was built: it explicitly describes translating/localizing an existing English-language Q&A source into Nepali, preserving the original meaning and facts — this is a localization dataset, not a from-scratch Nepali FAQ.

---

🌍 Important: This Is India-Focused Content, Written in Nepali

Despite being 100% Nepali-language, the subject matter is entirely about the Indian financial system — not Nepal's:

Term	Mentions (in answers)
भारत / भारतमा (India / in India)	93
नेपाल / नेप्से / NEPSE (Nepal / NEPSE)	0
सेबी (SEBI — India's securities regulator)	20
एनएसडीएल / सीडीएसएल (NSDL / CDSL — India's depositories)	2

Sample regulatory references found in answers include India's Finance Act 2024 and FY 2024–25 capital-gains tax changes, and demat-account rules tied to NSDL/CDSL — all India-specific institutions. If you're building a Nepal-market financial assistant, this dataset will not be a fit as-is; it's best understood as Indian investing content translated into Nepali, likely for a Nepali-speaking audience investing in Indian markets (consistent with "Lemonn" being an Indian investment platform).

---

💬 Conversation Structure

• Every record has exactly 2 turns: human → gpt. 300/300 consistent, no system prompt.
• Unlike the short, single-fact NEPSE dividend dataset or the pharmacy-exam reasoning dataset, this one has long, explanatory answers:

	Question	Answer
Word count (avg)	~5.6 words	~43.4 words
Character length (avg)	~32.1 characters	~288.4 characters
Min–max (chars)	17–57	153–449

Questions are short and punchy ("Is mutual fund income taxed?"), but answers are multi-sentence explanations with context, caveats, and reasoning — closer to a knowledge-base article snippet than a one-line FAQ answer.

Answer length distribution

Word count	Records	%
Under 30 words	5	1.7%
30–39 words	84	28.0%
40–49 words	158	52.7% ⭐ densest
50–59 words	46	15.3%
60+ words	7	2.3%

---

❓ Question Types

Questions cluster into a few natural shapes (not fixed templates — organically varied phrasing):

Type	Count	%	Example
Definition ("What is X?")	183	61.0%	"What is an AMC in a mutual fund?", "What is CAGR?"
How-to / mechanism ("How does X work?")	67	22.3%	"How does an IPO work?", "How does inflation affect investing?"
Yes/No	37	12.3%	"Can NRIs invest in mutual funds?", "Is SIP a safe investment?"
Quantity ("How much/many...?")	12	4.0%	"How much money do I need to start investing?", "How many stock exchanges are there in India?"
Other (e.g. "Why...?")	1	0.3%	"Why do people invest in stocks?"

The definition-style questions cover a lot of finance jargon/acronyms: AMC, AUM, CAGR, DRHP, ELSS, and more — this reads like a glossary-plus-FAQ hybrid.

---

🏷️ Topic Coverage (Keyword Frequency)

Keyword (Nepali)	Meaning	Mentions
कर	tax	127
म्युचुअल फन्ड	mutual fund	83
एसआईपी	SIP	38
इक्विटी	equity	38
आईपीओ	IPO	35
सेयर बजार	stock market	24
सेबी	SEBI	20
एएमसी	AMC (asset management company)	19
डिम्याट	demat (account)	17
स्टक एक्सचेन्ज	stock exchange	12
अवकाश	retirement	10
ऋण फन्ड	debt fund	7
पूँजीगत लाभ	capital gains	7
मुद्रास्फीति	inflation	6
एफ एन्ड ओ	F&O (futures & options)	5
ईएलएसएस	ELSS (tax-saving mutual fund)	5

Tax is by far the most recurring theme, followed by mutual funds, SIPs, and equity — consistent with a retail-investor-education FAQ set aimed at first-time investors navigating funds, IPOs, and tax implications.

---

🔁 Duplicates & Data Quality

Check	Result
Unique id	300 / 300 ✅
Unique questions	300 / 300 ✅ (no duplicates)
Unique answers	300 / 300 ✅ (no duplicates)
Turn structure violations	None (always human→gpt)
Stray Latin-script contamination	None found ✅ (unlike the pharmacy-exam dataset)
Missing/empty fields	None besides the intentionally empty url

This is the cleanest of the three datasets reviewed so far — no duplicate answers, no copy-paste mismatches, and no garbled text despite english_content_allowed: false.

⚠️ The one real quality flag: license: unknown

Unlike the NEPSE dataset (Apache-2.0) or the pharmacy dataset (CC-BY), this dataset's license and license_tier fields are both explicitly set to "unknown". Since the content also appears to originate from a commercial platform (Lemonn), this is worth resolving before any reuse — especially commercial use — since permissive-use rights aren't established the way they are for the other two datasets.

---

🎨 Design Philosophy (the likely thinking behind this dataset)

1. Localization, not fresh generation — the behavior_definition explicitly describes preserving an original question's meaning and an original answer's facts while rendering it in "pure Nepali," strongly suggesting this dataset is a Nepali translation/localization of an existing English (or Hindi) financial FAQ corpus from Lemonn, not natively-authored Nepali content.
2. Investor education, not fact lookup — the goal is clearly to teach retail-investor concepts (SIP, IPO, mutual funds, tax, demat accounts) rather than answer single numeric facts, which is why answers run to 40–50 words with context and caveats rather than one-line figures.
3. India-market grounding — all regulatory, tax, and institutional references (SEBI, NSDL/CDSL, Finance Act) are Indian, making this a India-focused product-education FAQ delivered in the Nepali language, not a Nepal-market dataset.
4. Broad glossary coverage — a large share of questions are "What is X?" style definitions of finance acronyms and jargon, suggesting this may be built to support a chatbot/help-center use case where new investors ask about unfamiliar terms.

---

🧪 Sample Records

{"id": "lemonn_001",
 "conversations": [
   {"from": "human", "value": "के म्युचुअल फन्डमा कर लाग्छ?"},
   {"from": "gpt", "value": "हो, भारतमा म्युचुअल फन्डबाट प्राप्त प्रतिफलमा कर लाग्छ। करको व्यवस्था म्युचुअल फन्डको प्रकार ... वित्त ऐन २०२४ ले आर्थिक वर्ष २०२४–२५ देखि लागू हुने केही पूँजीगत लाभ करका दरहरू पनि परिवर्तन गरेको छ।"}
 ]}

{"id": "lemonn_002",
 "conversations": [
   {"from": "human", "value": "के डिम्याट खाताबिना आईपीओमा आवेदन दिन सकिन्छ?"},
   {"from": "gpt", "value": "सकिँदैन। भारतमा आईपीओमा आवेदन दिन डिम्याट खाता अनिवार्य हुन्छ ... सूचीकृत धितोपत्रहरू एनएसडीएल वा सीडीएसएलअन्तर्गतका डिम्याट खातामा विद्युतीय रूपमा राखिन्छन्।"}
 ]}

---

⚠️ Things to Watch Out For When Using This Dataset

• License is unresolved — "unknown" on both license and license_tier. Confirm rights with Lemonn before any redistribution or commercial fine-tuning use.
• India-specific, not Nepal-specific — if the goal is a Nepal-market financial assistant, this dataset will actively mis-inform (wrong regulator, wrong tax regime, wrong depositories). It's suited for a Nepali-speaking-but-India-investing audience.
• Time-sensitive tax/regulatory content — references to "Finance Act 2024" and "FY 2024–25" capital-gains rates will go stale as Indian tax law changes in future budgets; these answers should be periodically re-verified against current rules.
• Long-form answers, not one-line facts — this dataset is better suited to teaching explanatory/educational response style than terse fact-retrieval, unlike the other two Nepali FAQ datasets reviewed.

---

📜 License

Unknown ⚠️ — not specified in the dataset; verify with the source (Lemonn) before reuse.

---

This README was compiled from a full programmatic analysis of all 300 records in the dataset (all figures calculated directly from the file, not estimated).