Category: Intelligent Content

Suppose your organization wants to become the preeminent source of information about a topic. It aims to give audiences the ability to look at any dimension of a topic they might be interested in. How would you offer this?

To deliver informationally rich content, numerous content items need to be associated to one another. Content needs to be modular, with components that work together. But how do these things relate to each other? Where does one start?

Content models define how units of content should interact. Content modelling can be difficult to grasp and practice, partly because it is not a single uniform method. It encompasses a spectrum of related approaches that can be adapted to different needs.

People sometimes start to model their content before they know all the content they really need. They focus on what content has been already created, and not explore what content is not yet available that might be of interests to users.

Content models are often more robust when they are backed by a domain model. A domain model enables content designers to untangle a messy topic and explore and define requirements and design solutions.

The role of content modelling

A content model is the end goal of a domain model. Rachel Lovinger has been instrumental in developing and advocating the practice of content modelling, so I will rely on her definition. She states: “A content model documents all the different types of content you will have for a given project. It contains detailed definitions of each content type’s elements and their relationships to each other.” She recommends using content models to bridge perspectives on a team.

“A content model helps clarify requirements and encourages collaboration between the designers, the developers creating the CMS, and the content creators.” — Rachel Lovinger

In addition to facilitating project delivery, content models improve how content is delivered to audiences. Content models can enable personalization, adaptive content, and content APIs. Cleve Gibbon, a collaborator of Rachel Lovinger in evangelizing content models, notes: “Great APIs are founded upon solid models. So if you’re building a Content API, be sure to create a content model FIRST that conveys the required level of structure and meaning.”

The spectrum of content modelling

Models can represent different dimensions of a topic: either conceptual, or formal and structural. Content models can indicate how to assemble content components. But first one needs to know how solidly your content types are defined.

On one end of the spectrum, you may have well defined, fixed content. In such cases, one can develop what Deane Barker calls a relational content model. He defines it as “the concept of how different, separately-managed pieces of content relate to each other.  (This is distinct from ‘discrete content modeling,’ which is how you structure a single piece of content.” He explains the goal as “the idea of taking multiple discrete content objects (articles, sections, issues) and ‘rolling them up’ into a more complex content object (publication).”

On the other end of the spectrum, you may have fluid content, where the exact requirements are still emerging and many different hubs of content are possible. In such cases, a domain focused, ontology based form of modelling can be helpful. This approach has been used by the BBC for several large projects. Mike Atherton emphasizes the importance of the domain of the topic in content models: “A content model maps our subject domain, not our website structure.” He advises: “Concentrate on modelling real (physical and metaphysical) things not web pages.”

One way to consider the differences in a content model and a domain model is the metadata they emphasize. Rachel Lovinger states: “The Content Model is primarily concerned with structural metadata, while the Domain Model is largely concerned with descriptive metadata.”

A domain model and content model are complementary. A domain model helps you describe things that will be represented by content, while the content model helps you structure the content. Using both allows you to understand the relationship of a real world entity with a content entity.

A domain model is a useful place to start when content does not yet exist, or one is looking for a fresh redesign of content. Domain models may be considered as the prequel to content models. By focusing on entities in the real world, and the relationships between these entities, one can see opportunities to develop content associated with these entities, and what elements would be needed for that content. The correspondence of domain entity type, and content type, is illustrated in the table.

Domain models in the real world: Italian wine

Domain models can clarify one’s understanding of a topic, and offer insights into how different items of information relate to each other. Domain modelling emerged as strategy in software development to bridge analysis and design of complex business domains by using a shared verbal and visual language between experts, endusers and developers. Domain models can be especially useful for complicated and messy topics. They would seem perfect for understanding Italian wine.

When you live in Italy, as I do, understanding Italian wine is a practical problem. Wine is ubiquitous, but understanding Italian wine is not self-evident. Walk into an Italian wine store and you are confronted with walls of bottles whose contents are largely unrecognizable. It’s not that all wine is difficult to understand. When I lived in New Zealand, I had a good idea what different wines were about. It’s Italian wine that is the challenge.

The famous wine critic Hugh Johnson once wrote: “the already bewildering complexity of Italian wines has become tangled enough to drive a critic to drink.” Italian wine is particularly hard to understand because of its heterogeneity. Even the imposition of standardized nomenclature to designate where a wine is from results in a bewildering array of non-standard implementations of these standards. Idiosyncratic traditions, politics, and rogue approaches mean that wines are described in great detail, but in richly differing ways.

At the core of why Italian wine is difficult to decipher is its product architecture: how specific wines are labelled. Consumers need an easy way to know the basic characteristics of a wine based on its label. Do consumers think of wine in terms of where it’s from (Burgundy) or what grapes it is made from (Chardonnay)?[1]  High volume wine producers have attempted to solve the product architecture problem by promoting brand awareness of a grape variety or a region. What happens when consumers are not familiar with either the origin name or the grape name?

Unfortunately, Italian wine labels are uncharacteristically difficult to decipher. Italian labels will show the producer + (grape variety and/or geographic indication) + year. That seems reasonable enough, until the consumer realizes that the only items on the label they might recognize are the digits of the year. Even if they have familiarity with another proper name on the label, that is not sufficient to make a selection decision.

The most significant piece of information about the kind of wine is indicated by the grape variety and/or the geographic indication (a regional designation similar to an appellation in France). Between these two items, there are nearly 1000 different varietals and zones that indicate the basic composition of the wine. [2]  To get a sense of how good the wine is, the most reliable information is the producer, and the year of vintage. Yet there are many thousands of wine producers in Italy of varying abilities, and the correlation of product quality to year of vintage is very specific to the variety of wine and where it was produced.

The complexity of Italian wine would seem tailored for digital content. But existing digital-only information sources on the web tend to be shallow — both in terms of their range of attributes, and their selective coverage.

Good information about wines, producers, and regions are available from several well known printed guides, such as those by L’Espresso, Gambero Rosso, Touring Club, Bibenda, and Slow Food. Despite the editorial quality of the content, the information is not as usable as it could be. Depending on the specific organization of the book, the information is stovepiped in one way or another. The editors of each guide assumes a fixed path of entry that generally leads to a producer profile. Users are expected to think like the editors to uncover information of interest to them.

In some cases there are iPad versions of these printed guides, but they don’t feel natively digital, and require lots of tapping to move around from screen to screen. They are less usable than the print version, because they are slower to move through, and one’s orientation can get lost when hopping between screens. The content, while structured editorially, is not structured digitally with digital metadata. There is no ability to move laterally through the content: navigation is hierarchical. Unfortunately shovelware that ports a printed product and dumps it into a tablet format is too common, due to the false promises embedded in Adobe InDesign.

What users need is not simply a catalog of items, but a way to make sense of the bigger picture, in addition to exploring the detail. The heavy focus on profiles means that the user doesn’t see easily how these items relate to other things. They also miss seeing collective behaviors of similar items, which is possible when one digitally aggregates items sharing the same metadata. Thinking through these relationships and behaviors is one benefit of domain modelling.

Understanding the domain

How do people think about a subject? Mike Atherton suggests: “Experts map the world, users mark points of interest.” It helps to know how experts think about a topic like wine, and then during design, figure out what more typical users consider high priority goals. What aspects of wine do people consider significant? How might different aspects be pulled together into interesting items of content?

The topic of wine is distinctive because many people want to become experts, in contrast to other products. Getting information about the product is rarely a perfunctory task, but a connoisseurial pastime. Some people want to develop a broad knowledge about all styles of wine, while other people want to have a deep knowledge about a few specific producers or product areas, perhaps tied to places they go on holiday. Many things people might be interested in are non-obvious. For example, soil characteristics can influence how a grape variety tastes. Others may be interested in the environmental credentials of a producer.

How to break things down so they can be managed

The most important task when developing a domain model is to identify appropriate entities. An entity is a thing, either tangible or conceptual, with a distinct identity. It’s not the same as an existing item of content — the content may not exist yet. Entities, to use the words of Cleve Gibbon, are “first class citizens in the business domain” — they are the actors in the drama on the stage.

Entities have attributes — characteristics. Attributes do not necessarily become a field in the content, but they often do. That decision needs to be made when the content is designed. Taste is certainly an attribute of wine, but is not necessarily a field in a description of a wine.

Once entities have been identified, it is necessary to determine where to put attributes, and whether to break entities into smaller units. Often, one discovers intermediate zones that straddle two entities. The horticultural characteristics of the vineyard reflect the interaction of the producer and the wine produced. The interplay between region and varietal defines the vintage for a given year. These intermediate areas may not deserve to be entities themselves, but one should consider how to make sure their role remains visible.

What a domain model for Italian wine looks like

It is helpful to first consider the relationships between entities, then examine the attributes associated with each entity.

When looking at entities, two things are important. First, how many instances are there for each entity type? The entity map shows that most of the entities, there are hundreds or even thousands of instances. This large number suggests that establishing meaningful relationships between entities will be important if users are to be successful navigating through such a large volume of content. Second, what is be essential character of relationships between entities? We want to know how many connections there are between entities: the more connections to other entities, the richer the potential interaction of information. We also want to know if the relationship between entities is a one-to-one relationship, a one-to-many relationship, or a many-to-many relationship. The “crow’s feet” in our entity map indicates numerous many-to-many relationships. That may make the design of content a bit more challenging, but it also indicates many interesting connections. Our content is a valuable resource when it’s not easy to see these connections in one’s head.

Next, explore the attributes associated with each entity. The goal is to identify and associate attributes of entities. Each entity has a number of attributes. Some will be short fields, others will involve longer text descriptions. There is no right number of attributes, provided all attributes are meaningful. The number of attributes to implement in design will depend on both business and design decisions. There will be a business decision concerning the cost of acquiring the information related to the attribute, and the usefulness to consumers of that information. There will also be a design decision relating to which attributes to expose to which audiences.

Our model shows attributes that are commonly associated with the domain of Italian wine. For example, it can be interesting to know the number of bottles produced of a wine. That can indicate how widely available the wine is to buy, or perhaps its scarcity (that one needs to reserve purchase). Some wine guides will indicate the total number of bottles according to producer, while others will indicate total number of bottles by label. This difference means that one can answer different questions, such as who is the largest producer within a geographic indication zone, or who is the largest producer of a specific kind of wine. Ideally, one would like data at both the producer and product levels, but that may not be easy to obtain for all producers.

Lessons from domain modelling

Even though domain modelling attempts to represent the real world, reality is often less orderly than we would like it to be.

Not everything can be easily expressed as a regularized attribute. Audiences will want to know: What does the wine taste like? It would be wonderful to provide a reliable, easy-to-understand way to explain taste that allows easy comparison between wines, zones, and producers. Sadly, taste is — surprise — a bit subjective. Different experts will say different things about the same wine, even when they agree on an overall judgment. Terminology is not standard either. The same words can mean different things. Critics may use the word “cherry” to describe a taste as “spicy black cherry” or as “cherry rhubarb.” There is no controlled vocabulary for wine, no limited set of descriptors with precisely defined and agreed meanings.

By their nature, models simplify reality. The geographic indication signifies where a wine is made, and the criteria by which it is made. Whereas most geographical entities are based on either political administrative geography or physical geography, geographic indications exist outside these frameworks. A geographic indication can straddle two administrative regions. It can exist in two different, discontinuous locations. Some geographic indication zones have subzones. Wine producers also can behave in complex ways. Sometimes a wine producer is a brand “house” that has vineyards in several locations, or a consortium that sources from different vineyards. The informational details associated with these exceptions may not be important to users, and can add design complexity.

The identity of items can be constructed in several ways. One needs to be able to distinguish one entity from others belonging to the same entity type — items need to be uniquely identified. Despite the challenges of deciphering Italian wine, specific entities fortunately are identified with meaningful, human readable names, rather than numeric product codes. The domain model can use existing identifiers, which are based on several approaches:

• Collectively defined names (the names of regions, geographic indications, and grape varieties), though some producers use alternate names for grape varieties.
• Self described (the name of producer), though sometimes producers choose to use both a house and proprietor name
• Inherited identity (the environmental profile for a producer)
• Names composed of compound attributes , such as dry sparkling rosato as a wine category entity.

Thinking about design

The domain model can support early design discussions. Many questions that are interesting to audiences will span two or more different entities. For example:

• What year produced the best wine from a region?
• What geographic indication commands the highest average prices?
• What grape varieties produce the most wine?
• What wines for a given year and geographic designation are ready to drink?

Some answers require computations of structured data. Questions of interest to audiences need to be translated into content types that will be represented in the content model.

In addition to supporting interesting exploration, the design needs to support common tasks. The domain model helps to identify information available to support common tasks. Some common points of entry audiences will seek when exploring wine include:

• By rating
• By price
• By category
• By variety

Users often focus on one specific criteria when starting the process of seeking information. In some cases, these are entities, in others, these are attributes. Considering task starting points can help identify potential groupings of content elements. Depending on the depth of content, these groupings may not be manageable for users without providing additional parameters to narrow the pool of candidate content. The most salient criteria is not the only factor that’s important to the user.

In contrast to starting points, another perspective is to consider the end goal of the task. Examining the end goal, the content designer can consider the orientation of different users. Users of wine information may be:

• Bottle centric — interested in the characteristics of specific bottles of wine
• Producer centric — interested in the story of the producer, perhaps with an intention of visiting them
• Food centric — mostly interested in wine styles as a complement to food dishes.

Domain depth and domain scope

The depth of a domain reflects both the number of attributes for an entity type, and quantity of items. Both aspects can impact the design. The quantity of items will influence content types that presents lists and links. The number of attributes will impact content type structures for content items.

Content designers decide how much of the domain model to present to users. A fixed content type may show all attributes as part of in content type. With a flexible content type, attributes may be optionally available, or have serval variations. Designers may choose progressive disclosure of content that hides details, which are revealed only when wanted. Or they may implement an adaptive approach, where different variations of content types are shown depending on the interests of an audience segment, or device formats.

The other aspect of the domain model, thus far unmentioned, is how it might connect with other domains. The domain model offers the possibility of enlarging the scope addressed by considering related domains. Different variations of content may draw on common content, while including different content as well (see diagram). Three different apps may share common core content. But they provide different functionality depending on their focus (touring vineyards, pairing wine with food, or knowledge enhancement of wine). The domain model can also be used to guide the planning of releases of content and functionality.

Relating entities: Comparisons to other approaches

Domain modelling is not the only approach to sorting through complex content. Before closing this discussion, it is worth talking about two other well known approaches that look and behavior similarly, but have some differences.

Faceted search, an approach popular in library science and information architecture, allows users to locate specific content by filtering on facets. Facets can be attributes or entities. The idea is that users can locate content that has the qualities of A & B & C. Faceted search is a popular technique, common on ecommerce sites, and is often helpful. The utility of the technique rests on several assumptions. First, faceted search assumes users know the two to four most important criteria, and will get a manageable set of results. If the set of results is large, users generally take a satisficing approach, happy with the first result encountered that is minimally acceptable. Second, faceted search presumes that each facet is independent of each other, which in the case of wine isn’t true. It is possible to get null sets if facets aren’t deep. While faceted search has been implemented on some wine ecommerce sites, it is not an effective approach for helping users discover content they might be interested in but not know about, and tends to focus on a limited range of aspects.

Linked data is an approach to modelling content that has close associations to domain modelling, thanks to the BBC’s integration of the two approaches. To simplify, linked data allows users to find content with characteristic A that has B, which has C. Organizing content using a linked data approach has both benefits and drawbacks. One drawback is that queries can be path dependent. Whether results appear promising or discouraging depends on how you construct the query. Linked data queries are generally more open ended than predefined structured queries that answer fixed questions with predictable sets of results. A bigger concern is that linked data treats all aspects of an entity as other entities, and each entity gets its own page. But not all attributes are meaningful entities — things worthy of their own content destination. On the positive side, linked data is good for what-else questions. One can link outside of a domain to other domains, such as to geophysical data.

Model behavior

Models aren’t reality, according to the cliche. Domain models may appear esoteric to some people, given that they aren’t actually something implemented directly, but are an input to other deliverables. To get buy-in for domain models, it may be best to use it as a discussion document, and note that it will evolve into the content model. While it l lacks the appeal of being code-ready, a domain model can play an important role on a project. It can uncover hidden requirements and opportunities, help align different stakeholders around a common vision, and accelerate the design process.

— Michael Andrews