The good news about free and open-source solutions for text analytics is that there’s a ton of them.
The bad news is that you’ll need a linguist working together with a data scientist to get some of them to work. “Some assembly required” is definitely true of many solutions for text analytics and sentiment analytics.
(Fiero parts, painted, on trees by Bill Abbott is licensed under CC BY 2.0)
For this reason, we’re focusing on tools that a normal business user can actually get up and running within a few minutes. We promise, you won’t need to compile source code or master complex algorithms. You may need to watch a few YouTube demos, but you were probably expecting that.
I’ve personally demoed the following solutions to test their ease of use. Many tools were demoed, but few were selected. Here are the ones we’ll cover:
(Click on a link below to jump to that section.)
RapidMiner + AYLIEN
Deployment: Windows, Mac, Linux
Works for: Sentiment analysis, advanced text analytics
Pricing: Free (free version processes up to 10,000 rows with a single logical processor—more advanced capabilities require paid version)
RapidMiner is a free, open-source platform for data science, including data mining, text mining, predictive analytics etc. The features of RapidMiner can be significantly enhanced with add-ons or extensions, many of which are also available for free.
Thomas Ott, marketing data scientist at RapidMiner, explains, “The beauty of RapidMiner is that it’s visual programming: You don’t have to write the code, and you don’t have to know the math behind it.”
Among other extensions, the RapidMiner Marketplace offers a very functional and user-friendly add-on for sentiment analytics developed by third-party vendor AYLIEN.
AYLIEN’s extension can automatically scrape data from Twitter (as can RapidMiner). It then analyzes tweets and scores them with a three-value sentiment scale: positive, negative or neutral.
In addition to reading from web sources such as Twitter, RapidMiner can also read directly from flat files such as CSV and Excel files or databases.
RapidMiner also offers its own extension for text analytics, which includes powerful text processing features that can be combined with advanced clustering algorithms and machine learning operators.
As Ott explains, “there are two main approaches to looking at text. One is doing a high-level overview: word counts, word frequency, where words occur in the corpus [the collection of documents being analyzed] etc. The other is more heavy-duty, e.g., sentiment analysis and other techniques in which you train a machine-learning algorithm on a data set.”
As you can see in the above screenshot, adding advanced analytics to a basic text mining workflow in RapidMiner is as simple as dragging and dropping operators into the proper locations.
Once this is done, it’s possible to output complex visualizations. For example, you can create a network showing the relationships between a specific term you want to focus on (such as a brand name) and other terms in the document you’re analyzing.
The following screenshot is an example of this kind of visualization. Ott applied clustering algorithms to Federal Reserve Bank meeting minutes to understand relationships between the currencies and concepts under discussion in the meetings:
When combined with sentiment analysis, such clustering techniques can have transformative impacts on small businesses in traditional industries such as retail. Ott gives a great example:
“I’m a beer brewer, and did some Twitter analysis of the brands that people are talking about based on region. It turns out that people on the coasts talk about IPAs, people in the midwest talk about stouts, and people in the southwest talk about ales. This is a key thing for small businesses to look at. Say, for instance, that I’m a Kia dealer, and I find out that people in Michigan like red cars and people in Montana like blue—I can then adjust my stock accordingly.”
Thomas Ott, marketing data scientist at RapidMiner
Takeaway: RapidMiner is the easiest to use and most fully featured text mining tool of the platforms I demoed. With the AYLIEN extension, you’ll be able to perform basic sentiment analysis within minutes of downloading and installing.
KNIME Analytics Platform
Deployment: Windows, Mac, Linux (cloud version also available)
Works for: Advanced text analytics
Pricing: Free (free version lacks advanced capabilities such as batch processing and template sharing, as well as advanced support options)
KNIME is another robust open-source data mining platform available in a free version with rich functionality.
Like RapidMiner, KNIME offers an intuitive visual workflow builder for “programming-free” data mining. It also offers a number of the same operators as RapidMiner (operators are known as “nodes” in KNIME).
With KNIME, users can perform tasks such as:
- Stemming: Collapsing variations on key terms into basic forms
- Stop word filtering: Removal of insignificant terms such as “in,” “for” and “the”
- Tokenization: Breaking text strings into smaller units, e.g., words and phrases via user-specified and pre-programmed rules
KNIME is also comparable to RapidMiner in its ability to read directly from Twitter as well as flat files such as CSVs.
I personally found KNIME’s workflow interface to be somewhat more difficult to use than RapidMiner’s, despite their similarities. For example, the logic of the input/output pipeline strikes me as being implemented more naturally in RapidMiner than in KNIME.
Additionally, RapidMiner offers detailed automated suggestions around why operators in your workflow don’t connect, which make it easy for a complete novice to build a functional text mining pipeline and troubleshoot problems when they occur.
KNIME offers good descriptions of nodes, but explanations of why nodes won’t connect are often cryptic.
Finally, RapidMiner currently offers more text processing and sentiment analytics extensions than KNIME.
On the other hand, the free version of KNIME offers more extensive data processing capabilities than the free version of RapidMiner. There’s no limit to the number of rows you can process with the free version of KNIME, making it better for large data sets, or on the number of physical cores/logical processors the free version can leverage in data processing.
Takeaway: Experienced business analysts and data scientists will be comfortable using either RapidMiner or KNIME, and should demo both in order to make a decision based on the advanced functionality of these platforms. Novices will be better served by starting with RapidMiner. Not only is RapidMiner’s interface easier to learn, but there’s also more documentation out there on how to use it.
Open Calais (now Refinitiv)
Deployment: Cloud (on-premise version is paid)
Works for: Entity recognition
Pricing: Free (limited to 5,000 submissions per day, whereas paid options scale into the millions and offer more extensive sets of category fields for tagging)
Open Calais (now Refinitiv) is a cloud-based content tagging tool offered by Thomson Reuters. Unlike RapidMiner and KNIME, it’s not a data mining suite with text mining extensions, and it doesn’t do sentiment analysis. Instead, it excels in the realm of entity recognition and extraction.
You feed unstructured text into the tool, and it recognizes entities such as people, products and companies. It also recognizes relationships between entities and facts about entities. It even organizes entities into topics.
Open Calais (now Refinitiv) can thus be used to quickly extract information from documents. This information can then be used to tag documents for classification.
Some use cases for this functionality include:
- Tagging blog articles to improve navigation on a site
- Tagging internal resources on a corporate intranet to help employees find them using search
- Tagging knowledge base articles and academic archives etc.
Takeaway: Unlike RapidMiner and KNIME, Open Calais (now Refinitiv) won’t work for basic text processing or advanced sentiment analytics. It’s very good at recognizing entities for analysis of unstructured text, and is a robust tool for document tagging.
Deployment: Windows, Mac, Linux
Works for: Determining term frequency, vocabulary profiling
AntWordProfiler is a freeware tool created by Laurence Anthony, a professor at Waseda University’s Center for the English Language Education in Science and Engineering. Anthony has a PhD in linguistics, and the tool he’s created excels at quick vocabulary profiling of large files.
AntWordProfiler uses preloaded vocabulary and thesaurus lists, which can be edited by the user, in order to determine word frequency. Users can also load custom vocabulary lists into the tool.
Results can then be saved in a text file formatted for easy viewing in Excel or another spreadsheet tool. There’s also a document viewer that highlights where terms in your vocabulary lists appear in the document.
Takeaway: AntWordProfiler can be used for quick counts of word frequency in complex, unstructured texts, as well as custom vocabulary profiling of unstructured texts. Unlike RapidMiner and KNIME, however, it’s not an end-to-end text mining solution.
Grab Bag: Even More Toys!
Here are a few other neat toys you should consider experimenting with:
Carrot2: A dedicated tool for applying clustering algorithms to documents. There’s a web-based interface for applying some common clustering algorithms that can help with organizing documents into thematic categories. Carrot2 also integrates with the APIs of popular search engines in order to automatically cluster the results of keyword searches. It can thus be used in search engine optimization (SEO).
AYLIEN Google Sheets add-on: AYLIEN, the same company that develops the sentiment analytics extension for RapidMiner that we examined, also offers an add-on for doing sentiment analysis directly within Google Sheets. This is one of the easiest ways to score sentiment in a spreadsheet-style interface, but the number of API calls you can make per day with the free plan is limited.
National Centre for Text Mining/University of Manchester Sentiment Analysis: While still in beta, this tool is already quite functional in determining the overall sentiment in a single text (batch upload isn’t supported at this point). One nice feature is that the tool highlights positive and negative terms and chunks of text in different colors.
The Data Science Toolkit: A collection of easy-to-use, web-based text mining tools, including basic sentiment analysis. The sentiment analysis tool only supports analysis of short chunks of text at this point. There are also lots of tools for geocoding text. For instance, you can translate street addresses to coordinates. These tools are also available via API calls for advanced use cases.
Proprietary BI Platforms: If you’re beginning to feel like the free stuff won’t work for you and would like to begin exploring the text mining features of paid BI solutions, you can examine user reviews and ratings on our site.
If there are additional open-source and free text mining tools you think we should list here, please just drop me an email at email@example.com.