Q&A

Welcome to our API Service Q&A section. Here, you'll find answers to common questions regarding our API access, usage guidelines, technical details, and compliance information. If you have additional questions or need further assistance, feel free to reach out to our support team.

---

Sources

We started quite small for a reason, this being to keep it condensed enough so we can control it. We’ve been more focused on quality data. We realized that it is better if we have fewer sources but they are more relevant. Nevertheless, we are able to push in a new source relatively quickly so do get in touch if you find that one is missing.

Absolutely, especially for tech-related queries. We're a small team, so we can't promise immediate changes, but we'll prioritize your request and provide updates.

New sources are integrated into metrics after a calibration period to ensure accurate representation in our data analytics. This means that it may take time to show up in reports. You can expect the daily report to have a new source two days after it has been introduced and 2 weeks for the weekly report.

Currently, our focus is on tech-related content. Plans to expand into other areas depend on further developments and demand.

---

Metrics

Count means the amount of times a keyword was found in the data for the period. We only count a keyword once per text.

The models that extract keywords may extract different versions, such as it may extract the full word as AWS Glue or it may only extract Glue. People may also talk in different ways about a keyword, such as Zuckerberg may be called Zuck, X will be called Xitter or ECS may be called Elastic Container Service. We do have a system in place that does try to recognize and normalize keywords but it's not perfect. So, count may not be 100 correct as in it may miss a small positive portion of data but it does not lie to the data that it does find. So if a count is high, it may be higher but it won't be lower.

We have an algorithm in place that calculated this based on other metrics and historical data. You may sometimes see a keyword that is trending yet it has not increased in count from the previous day, in these cases it is not mistaken but looks at other parameters.

No it is not. It is a work in progress and we hope we'll build better models in the future as it is now it does well to recognize shifts in data but we would need better processes to recognize keywords that are making less noise.

The engagement metrics is a bit diffuse but very important. It's how we try to establish authority of a post. Sometimes, we estimate a number based on previous metrics and sometimes we're able to pick up data from the page which will tell us how people are engaging with the content. This is not always something we can do, and it won't always be correct. Sometimes a post or an article will have more engagement than what has been estimated here.

Sentiment is context based which means it is set at the text level and not the keyword level. This means that if a keyword is mentioned in a comment that has been analyzed as negative, the keyword will be set as negative for this specific text. When looking at sentiment you should look at the number of texts for the keyword with that particular sentiment to make any kind of decision on what it happening. You should also look into the sources to see what people are saying to understand in what way a keyword is seen as "negative" or "positive".

We do though see that if a majority of texts are negative, people are talking negatively about a keyword.

---

Categories for keywords

We have categorized the keywords as following: Companies & Organizations, Tools & Services, Platforms & Search Engines, Hardware & Systems, Frameworks & Libraries, Languages & Syntax, AI Models & Assistants, Websites & Applications, People, Subjects, Concepts & Methods, and Bucket (other).

This category is usually excluded from reports altogether. It may contain everything that we have sorted as irrelevant along with words that are difficult to categorize. We recommend excluding it but it may be interesting to look into it once in awhile. Think of it like your email's spam filter.

You've touched on something important. A keyword can have multiple meanings. For instance, sometimes people are talking about Celery, the vegetable, and sometimes they are talking about the tool. The model understands context so it should correctly categorize Celery, the vegetable, as Bucket (other) and the tool Celery as Tools & Services.

As with any AI system, it is never 100% correct. We do see mistakes happen but the models are being rebuilt continously to improve their outputs. We see that the model may have issues to distinguish a new word it hasn't seen before but does well 95% of the time.

---

Access and Costs

We provide free access to public endpoints with burst and throttle limits. For extensive use, you can obtain an API key for private endpoints with daily unlimited access.

---

Technical Details

We use NLP models to extract the most relevant keywords from texts. These models identify key terms that signify the main topics of discussion.

Our sources endpoint is designed to be able to handle search, allowing you to find all related mentions across our data sources. This is a great way to do research on a subject and find relevant sources.

While our system aims for accuracy, occasional dips may occur due to missed crawls. We continuously refine our crawling processes to capture data as comprehensively as possible.

---

Legal

Yes, our service operates legally as a search engine, focusing on metadata and analytics rather than copying full texts. If you want to read the entire article you always navigate to the source with the link provided with the sources.

---