Comments Holistic Open Platform: How to fix news comments

The Knight-Mozilla News Technology Partnership is seeking ideas on how to reinvent journalism for the open Web, so they’re running a series of challenges. The first Knight-Mozilla challenge focused on “unlocking video.” The second aims to solve the news comments problem by going beyond comment threads:

One of the best things about the web is that it enables many voices to be heard. Blogs, comment threads, forums, and social networks empower people to take part in new kinds of discussion, dialogue, and debate.

The best discussions around the web can be pretty isolated. Take comments, tweets, and other fragments out of their original context, and they can become meaningless. And take a look below the fold—in comment threads at news outlets, political blogs, YouTube, and elsewhere, you’ll often find that the loudest voices drown out everyone else.

Knight-Mozilla News Challenge - Beyond Comment Threads logo: AtomAt the same time, media is moving beyond the traditional “news story” as the only unit for commenting and interaction, stretching to include narrative arcs of multiple stories over periods of time, “explainers” that provide background knowledge for strings of stories, “streams” that include initial reports followed by updates and corrections, and more.

With all that activity happening across the web, how do we enable more coherent, elevated discussion? How can news organizations improve the signal-to-noise ratio in public news commentary?

Here’s my proposal:



This is a holistic approach to what I see as the five major problems facing comments today and in the near to medium term.

  1. QUALITY. The major problem in commenting on news sites today is finding a way to maintain high-quality discussions.
  2. DISCOVERY/PERSISTENCE. Commenting is fragmented and no system properly federates them.
  3. COMMENT FORMATS. No truly transmedia commenting system exists.
  4. CONTEXT AWARENESS. Commenting systems don’t address people’s needs in various device, temporal and physical contexts.
  5. EMERGENT MEDIA. There is no standard for comments in emergent media and platforms. Each element of this plan can be developed separately, or as part of a phased, holistic solution.




Maintaining high-quality discussions.

    a) Community ranking / moderation + TrustRankCommunity ranking’s fundamental flaw: It rewards popular ideas and unpopular ones are often submerged. To offset this, an algorithm could assign a TrustRank score that surfaces comments from trusted people on a sliding scale weighted by the viewer. TrustRank improves over time and volume of comments.
    b) Semantic / sentiment parsing e.g. SentimentRankCommunity or algorithmic ranking and/or parsing of meaning and sentiment would assign a score viewers could weight. SentimentRank would find commenters with a similar temperament and outlook and help determine which comments they see. This improves over time and volume of comments and participants.
    c) Crowdsourced or automated summaries of longer comment posts
    As comments accumulate, reviewing and understanding the discourse can become onerous. Volunteers or an algorithm could identify and summarize key themes for brief, headline-style summaries to help viewers discover and understand context.
    d) Aggregate summaries and sentiment scores into heatmap / graph / that surfaces key points
    Rankings can be visually displayed in a map or graph. This would help to surface key thematic comment clusters that the viewer could drill down on for finer granularity.
    e) Badges / incentives / credits or scrip toward paid services for commenting
    Commenters receive rank badges to enable viewers to quickly assess the quality of an individual comment in a historical context. This could be combined with credit or scrip system that news organizations can use to reward commenters toward paid services. The open platform would make rewards portable across outlets using it.
    f) Trust circles, connected communities on other networks, TrustRank + SentimentRank sort
    Subject to individual preferences, people see comments from friends in social networks, extended networks, then people outside their networks.
    g) Present commenter with others’ comments inconsistent with own views
    Combine factors to offer viewers comments that oppose their own, to help stimulate meaningful debate vs. a cargo cult.



Commenting is fragmented. While multiple platforms exist to federate comments, they still occur in isolated islands. To address this:

    a) Open standard that federates and categorizes comments
    Federate across services: blogs, status, chat, photo, video, text, SMS, etc. (Opt-in.)
    b) Visual / audio / tactile clustering
    Comment systems are heavily biased toward educated, literate, able-bodied individuals. Inclusive commenting systems would also assist the fully able. Standardized/automated markup would help identify and enable content federation for the multimedia and emerging sensory/haptic Web and enabled devices. People could navigate comment heatmaps parsed for subject, sentiment and trust through visuals, audio or tactile/force feedback. Similarly, content types could be toggled as comment options.
    c) Static vs. real-time
    Static comment threads are easy to hijack. Real-time commenting and discussion archived or parsed for inclusion in a historical forum/thread would enable actual discussions vs. turn-based commenting and repetitive or irrelevant crosstalk.



Transmedia comments vs. plain text.

    a) Text – Real-time commenting and discussion into archived/threaded forum enables collaboration with journalists on a story before publication-to-deadline, and fosters higher quality ongoing commentary and story development post-deadline. Granular word/sentence/paragraph level comments can be tagged.
    b) Images – Image-post comments for people with varying time, literacy and forms of self-expression. Character recognition and transmedia publication for any text within images, sentiment-parsing for facial expressions or gestures, and image summaries can be extracted and posted via community scoring or algorithm.
    c) Audio – Audio comments can be parsed as text and ranked for sentiment and trust.
    d) Video – Video comments (time delimited) can be parsed into text posted with and/or annotating video. To sort multiple video comments, they would be thumbnail stacks akin to BumpTop’s concept. Brushing a clustered stack would surface the video thumbnail on a card with the commenter’s profile, semantic TrustRank + SentimentRank score and/or graph, personality matching, and a capsule comment summary.



The type of comment a person can leave often depends on time of day or their physical location:

    Office / mobile: Likely fosters shorter comments.
    Home / tablet: Likely fosters longer comments.
    Sorting and serving up comments to viewers and commenters based on set preferences, time of day and physical contexts including Web-enabled objects could foster higher quality discussions by reducing the frustration factor for commenters and viewers.



    a) Projected/sensors: Pico-projectors, digital vision, tactile and other sensors can shift comments from a solitary experience to a shared one. Collaboration in physical space is essential.
    b) Wearable: Personal augmented reality and wearable displays demand that comment systems include the ability to interact in real-time and with geolocated and physical, Web-aware objects.