The quality of reader comments on news websites is varying a lot. It costs time to scan through the less interesting comments to find those who really add an interesting point of view, link to sources or contain valuable background information, which a news article not cover. Moreover readers tend to stay with a single online newspaper, therefore it's unlikely to be confronted with views of other commentators.
The quality of the reader comments on news websites is varying a lot. It costs time scanning through the less interesting comments to find those who really add an interesting point of view, contain valuable background information or link to sources. Moreover readers tend to stay with a single online newspaper, therefore it's unlikely to be confronted with views of other commentators.
A platform using machine learning technology to crawl qualitatively valuable reader comments from different news sites and presents them in dossiers.
A plattform algorithm based collection of valuable reader comments from news websites.
Scanning the enormous amount of comments and decide which are valuable, would be quite time consuming for a person and could be also affected by it personal opinion. A machine learning method is a more efficient alternative dealing with the big amount of data.
When using algorithms, it is important that the user understand why a comment is selected by an algorithm. To avoid blackbox algorithms, it is advisable to choose the LIME approach. LIME (Local Interpretable Model-agnostic Explanations) has the ambition to make every decision by the algorithm transparent, which increase trust in the algorithm.
The algorithm needs to be able to deal with different structured news websites in order to identify the comment section. An iterative, constantly improving algorithm devlopment is cruscial to maintain quality of collecting of comments. Powerful algorithms influence already our lifes today, e.g. in stock markets or for medical diagnosis purposes.
The platform presents reader comments in dossiers. The name of each dossier is build out of headline tags, which allow the user to grasp quickly the topic of a dossier. These headline tags are important because they holding the not necessarily directly related comments together. It's key to understand that the goal is not to generate a narrative within a dossier; a dossier provides a collection of comments to a certain topic area.
Within a dossier a user is able to recognize the amount of reader comments. For each comment the user sees the author and the news website, where that comment comes from. An indicator tag shows why a comment is valuable. An optional extended explanation provides more information on how the algorithm have worked for each indicator tag.
Guiding-transitions provide a intuitive user experience, where focus mostly lays on content perception.
The shape of the logo references to a typical speech bubble that is often used as a symbol for comments in general. It gives the user a quick idea what Crawier is all about. The name Crawier is a composing out of “crawling” and “dossier”. A hint to the concept of crawling news articles and collecting them in dossiers.
The contrast of a black and yellow combination evokes attention. A light green is used as the secondary highlight color, e.g. to highlight good parts of a comment or a positive indicator tag.
The Lekton typeface is a small reminiscence to the old typewriter journalism. The trispaced glyphs allowing a vertical alignment of single letters. Together with a increased letter-spacing, it creates a distinctive look for "Headline Tags" and the visual apprearence overall. The typeface Open Sans is highly legible on screen, even at small sizes, and sums up to an ideal body text.
Every headline-tag is based on newspaper sources e.g. puplished tags by an online-newspaper. It replaces the usually headline and allows a quick overview for each dossier. It is possible to adjust and thus personalise a dossier, just by editing headline tags.
The Indicator-tags give a quick idea about a given rank. The tags show only positive features of a comment - it makes it quick and easy to grasp the reason for a rank, especially because of the numerous possible tags itself. Clicking on a tag shows an extended explation.
The upcoming feature for a browser extension allows the reader to “support” the algorithm in finding valuable comments. Users can directly point out comments for the algorithm. That gives the algorithm the chance to check certain comments with a higher priority. Through this user-input the algorithm learns which comment is valueable.
Points of interest
This feature enables user to recognise quickly points of interest in comments. Which part of a text is highlighted is defined by different algorithm analysis methods.
Crawl — Analyse — Rank
The basic principle of the algorithm involves three major steps. First task is to scan certain news websites, and look out for the comment section and collect comments with the related information (i.e. user data, comment rating and text phrases). Then, certain parts of each comment are analysed by using different methods (e.g. text mining, natural language processing and sentiment analysis). The last step is to rank the comments and to cluster them into meaningful dossiers.
Crawl — Analyse — Rank
The basic principle of the algorithm involves three steps. First task is to scan certain news websites, and look out for the comment section and collect comments with the related information (i.e. user data, comment rating and text phrases). Then, certain parts of each comment are being analysed by using different methods (i.e. text mining, natural language processing and sentiment analysis). The last step is to rank the comments and to cluster them into meaningful dossiers.
concept, ui/ux, brand identity