### About this dataset

Introducing our revolutionary STEM Question-Answer dataset for tech companies building LLMs. Unlock the power of meticulously curated one-to-many question-answer pairs across domains like chemistry, math, astronomy, biology, computer science, and physics. The dataset is currently under collection and will feature 150,000 units in English.

Enhance understanding, reasoning, and problem-solving capabilities of your LLMs with our comprehensive dataset. Join us today to revolutionize your tech company's LLM development and achieve unprecedented accuracy and proficiency.

This dataset is covered by our standard [Data License Agreement](https://marketplace.defined.ai/dataset/data-license-agreement). The license agreement is perpetual and allows for the commercialization of all models built on the data.

### Samples
#### Preview
![STEM_LLA_Sample_Json_Preview.png](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/STEM_LLA_Sample_Json_Preview_7d285cf413.png)

Tell us about yourself, and download samples of this dataset

#### Full Data Sample
Download free full data sample [here](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/STEM_llm_data_sample_480fcbd049.rar)

Download Sample

Large Language Models Data

Healthcare Prompt and Response Data

Aspect-Based Sentiment Analysis

Speech

English Spontaneous Dialogue

Named Entity Recognition

You might also be interested in:

STEM Q&A Pairs

### About this dataset

This LLM dataset can serve as a catalyst for AI teams building models for healthcare. This meticulously curated collection comprises hundreds of thousands of real-world physician prompts and their corresponding machine-generated responses.

Through collaboration with general practitioners (primary care doctors) across four markets, we have cultivated a diverse and representative dataset. The dataset includes both clinical and non-clinical conversations, segmented by various vertices such as country, region, specialty, gender, and age group.

The dataset does not contain any personally identifiable information or patient data.

Q&A pairs are available in English (US), Italian, French (France), Portuguese (Portugal) and Spanish (Spain) languages.

This dataset is covered by our standard [Data License Agreement](https://marketplace.defined.ai/dataset/data-license-agreement). The license agreement is perpetual and allows for the commercialization of all models built on the data.

### Samples
#### Preview
![Healthcare_QA_LLM_Sample_3.png](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Healthcare_QA_LLM_Sample_2_json_aff8116136.png)


50k+ annotated questions and answers pairs

## About the Dataset
### 3345 hours

This audio dataset contains 3345 hours of English Speech Data in various domains, recorded by native speakers from the UK, the US, Ireland, Australia, and India.

There are **1309 hours of English (US)** human-to-human audio with the following domain distribution per dataset:
- 321.37 hours of Banking
- 192.03 hours of Insurance
- 51.32 hours of Retail
- 45.68 hours of Telecommunication
- 198.62 hours of Generic domain
- 500.22 hours of Healthcare / Retail

There are **996 hours of English (UK)** human-to-human audio with the following domain distribution per dataset:
- 225.53 hours of Banking
- 196.05 hours of Insurance
- 189.67 hours of Retail
- 203.67 hours of Telecommunication
- 181.38 hours of Generic domain

There are **702 hours of English (India)** human-to-human audio with the following domain distribution per dataset:
- 173.55 hours of Banking
- 187.08 hours of Insurance
- 168.62 hours of Retail
- 173.4 hours of Telecommunication

There are **304 hours of English (AU)** human-to-human audio with the following domain distribution per dataset:
- 86.03 hours of Banking
- 76.35 hours of Insurance
- 79.37 hours of Retail
- 62.73 hours of Telecommunication

There are **34 hours of English (Ireland)** human-to-human audio with the following domain distribution per dataset:
- 28.3 hours of Insurance
- 6.58 hours of Generic domain

Defined.ai creates scenarios for our crowd members to follow, which they study beforehand. They then record a conversation, one speaker playing the agent, the other speaker “playing out” the scenario with spontaneous content. The recording is done via telephony and is saved in 8khz 16 bit per channel. That content is then transcribed.

The dataset is covered by [Defined.ai's standard license agreement](https://marketplace.defined.ai/dataset/data-licence-agreement). The license agreement is perpetual and allows for the commercialization of all models built on the data.

Other characteristics:

- Audio format: WAV
- Recording environment: noisy, silent
- Bits per sample: 16
- Communication band: broadband
- Sample rate: 8Hz

## Metadata Distribution
### Australia
![Spontaneous_En_Au_Gender.png](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_En_Au_Gender_91b267a1cc.png)
![Spontaneous_En_Au_Age.png](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_En_Au_Age_60e48eeca6.png)
![Spontaneous_En_Au_Accents.png](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_En_Au_Accents_e683db2545.png)

## Short Audio Samples
- [English (UK)](https://defineddata.blob.core.windows.net/samples/AFI_en-gb_multi-spontaneous_banking_30m_v01_Sample/Audio/a2d24fc5-da9d-4ffe-aba3-5b73dccf5237.wav?se=2024-06-15T16%3A49%3A44Z&sp=r&sv=2020-06-12&ss=b&srt=o&sig=a%2B2BmW0bflObwTrc0Xli6Blg/yyKxqrCwyqNkg9rWeo%3D). Transcription for the sample is also [available](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_English_UK_Banking_Short_Transcription_42d5aaed87.tsv)
- [English (India)](https://defineddata.blob.core.windows.net/samples/AFI_en-in_multi-spontaneous_banking_30m_v01_Sample/Audio/5bb7a3eb-db5e-4cc0-a3d2-6f11f00711b2.wav?se=2024-07-13T22%3A32%3A34Z&sp=r&sv=2020-06-12&ss=b&srt=o&sig=mZy7yOT4039to6Uqr3RE4R8XjhnWZ6sAZ2EiaJx/0sI%3D). Transcription for the sample is also [available](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_English_India_Short_Transcription_679275bf52.txt?updated_at=2023-06-06T15:22:23.708Z)
- [English (AU)](https://defineddata.blob.core.windows.net/samples/AFI_en-au_multi-spontaneous_banking_30m_v01_Sample/Audio/dd8bfb26-003a-4da5-ba07-5acde431b0c0.wav?se=2024-06-15T15%3A20%3A19Z&sp=r&sv=2020-06-12&ss=b&srt=o&sig=wiIvr8sC5QPZh77G%2Bl0D40XaaOwBft7Dcui66MA3aFI%3D). Transcription for the sample is also [available](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_English_Australia_Short_Transcription_7301ed0f39.tsv?updated_at=2023-06-06T14:54:56.892Z)
- [English (Ireland)](https://defineddata.blob.core.windows.net/samples/AFI_en-ie_multi-spontaneous_insurance_B1_v01_30m_Sample/Audio/afe8facd-9c34-48fe-a2a5-7d8ce377807b.wav?se=2025-07-06T13%3A12%3A14Z&sp=r&sv=2020-06-12&ss=b&srt=o&sig=ZY%2BhyzDxhmtJG80KR0lCLwOPVjVK2Z9RW3IyM4AXf2Y%3D). Transcription for the sample is also [available](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_English_Ireland_Short_Transcription_877343e4a7.tsv?updated_at=2023-06-06T15:26:39.452Z)
- [English (US)](https://defineddata.blob.core.windows.net/samples/JPT_en-us_multi-spontaneous_banking_30m_v01_Sample/Audio/b213bc4f-4f61-463e-837a-fe6bd00df5f9.wav?se=2024-06-15T22%3A49%3A38Z&sp=r&sv=2020-06-12&ss=b&srt=o&sig=zneqq2xQ73KK9l9ehzqVO4qdfj8er56vclSMhJ7HDK8%3D). Transcription for the sample is also [available](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Spontaneous_English_US_Banking_Short_Transcription_b667de8c76.tsv)




3345 hours recorded by speakers from the UK, the US, Ireland, Australia, and India

## About this dataset

Our Aspect-Based Sentiment Analysis (ABSA) datasets are the next step up from your standard sentiment annotation dataset. Instead of one annotation at the sentence level, these datasets indicate aspects and opinions at the words or phrasal level of product and service reviews, as well as the relationships between each. 

With these datasets, you’ll be able to train models to identify which opinions are expressed on features of products or services, or to find complex correlation patterns between opinions and features or other data points – in other words, you get to know in detail what your customers really think!

The dataset features 60,000 units in Japanese, Mandarin, Spanish, German and English (US).

## License Information

This dataset is covered by Defined.ai standard [Data license agreement](https://marketplace.defined.ai/dataset/data-license-agreement). The license agreement is perpetual and allows for the commercialization of all models built on the data.

## Samples
### Preview
![ABSA_EN_Short_Sample_PH1_PH2.png](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/ABSA_EN_Short_Sample_PH_1_PH_2_e4e0ef7e04.png)

### About this dataset

You get access to 24 categories of annotated named entities, ranging from the typical person names, locations, and company names to markers for date, time, and duration - amongst many others. Train models to be able to identify any entity relevant to your chatbot or NLP application!

The dataset features 150,000 sentences in Norwegian (Bokmal), Finnish, Turkish, Hindi, Arabic, Danish, Swedish, Hebrew, Russian, and Czech.

### License Information

This dataset is covered by Defined.ai standard [Data license agreement](https://marketplace.defined.ai/dataset/data-license-agreement). The license agreement is perpetual and allows for the commercialization of all models built on the data.

### Sample
![NER_RU_Short_Sample.PNG](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/NER_RU_Short_Sample_5e97248386.PNG)

STEM Question-Answer Dataset of 150,000 units coming soon