About this book Hierarchy is a form of organisation of complex systems that rely on or produce a strong differentiation in capacity power and size between the parts of the system. Show all. From the reviews: "This book, consisting of seven chapters from eight authors in addition to editor Denise Pumain, looks at the nature of hierarchy in a variety of disciplines. Conclusion Pages Pumain, Denise. Show next xx. Read this book on SpringerLink.
Recommended for you. When engaging in research, they also use a variety of data collection techniques that include objective measurement of performance or physiological indices; subjective ratings of satisfaction, comfort, or workload; observational checklists; and interview methods. As with any discipline, the research method chosen and the type of data collected depend on the nature of the problem and other issues, such as feasibility, cost, and time constraints.
For example, in order to design cognitive aids to support the ability of people to engage in Internet-based health information seeking, it is important to understand the cognitive abilities that are required to perform this task successfully. This would typically involve conducting research in a laboratory setting to investigate the relationship between cognition. The research protocol is likely to entail assessing the cognitive abilities of study participants using standard measures of cognition; having the participants perform a sample set of health information search tasks; asking the participants to rate the level of difficulty of the tasks and identify the sources of difficulty; and examining the relationship between the measures of cognition and measures of performance e.
Another example is an observational study, in which the goal is to understand if the prevalence of Internet-based health information seeking varies among age or ethnic subgroups. In this case, telephone or mail surveys or real-time tracking of Internet behavior might be used to gather the needed information. Sometimes the information gathered about behavior is used to develop mathematical models or simulations, which can then be used in the design of tasks or technologies.
For example, biomechanical models are often used to evaluate or compare physical demands of tasks or environments. These models might be used to predict the amount of stress on the spines of caregivers to help in the design or selection of mechanical aids for transferring care recipients from a bed to a wheelchair or shower. Optimally, human factors methods and principles are involved in all aspects of the design process, including the predesign analysis, design expression and prototyping, testing, and evaluation see Figure Food and Drug Administration for some medical devices on the market.
Human factors specialists use a variety of methods to support the design process. The overriding principle is to center the design process on the person or persons in the system; in other words, human factors practitioners adopt a user-centered design approach. User-centered design, as a design philosophy, has been around for several decades e.
- Britain and the Netherlands in Europe and Asia: Papers delivered to the Third Anglo-Dutch Historical Conference.
- You are here.
- Hierarchical networks of scientific journals | Palgrave Communications.
- Queer Tracks: Subversive Strategies in Rock and Pop Music?
- Probability. An introduction.
In fact, user-centered design has been elevated to a standard International Organization for Standardization, [ISO ]. As needs are determined and design features conceptualized, it is useful to develop prototypes of the design at each point in its development and to test these prototypes with the intended user population. Often the information gained from such prototyping and testing will need to be fed back to inform changes in design that will improve the product or system. This repeated prototyping, testing, and revisiting of the design, shown by the recursive arrows in Figure , is the best way to ensure good fit with user needs, expectations, and capabilities.
Even after the product or system is marketed, it is useful to solicit and analyze feedback on it from users to inform updates or new designs. The user-needs analysis usually includes such characteristics as age, education, gender, culture and ethnicity, physical and cognitive abilities, relevant skills, language, and literacy, among others. Gathering this information might involve conducting interviews with potential users to understand their goals and objectives with respect to a particular system or system component, such as a device, where it will be used, how often it will be used, experiences with similar devices, etc.
It is important to recognize that, for health care in the home, the users are heterogeneous and include people who engage in self-care or receive care and both lay and professional caregivers who vary widely in their skills, abilities, and characteristics see Chapter 2. As noted in Figure , the environment is multifaceted and not restricted to its physical characteristics. It is also used in the evaluation of existing systems to help identify design problems and sources of mismatches between system demands and user-group capabilities. These demands are then compared with the capabilities of the planned user population to determine where errors and inefficiencies are likely to arise.
The result is a list of potential mismatches keyed to each task and subtask, which is the basis for deriving design requirements for a usable system. The current standard task analysis methodology is hierarchical task analysis HTA , although many methods are available. HTA starts from system goals and uses a systematic goal decomposition methodology until a sufficient level of detail is reached to solve the problem at hand.
The result of the analysis is generally a hierarchical structure that can be represented either graphically or in an outline-like formatted table that organizes tasks as sets of actions used to accomplish higher level goals. Chapter 4 presents several examples of this methodology. In health care, many tasks, especially those relying on the use of technology, draw heavily on cognitive capabilities with users required to receive, understand, evaluate, and act on information.
For these, one might perform a cognitive task analysis, which can be conceptualized as task analysis applied to the cognitive domain. In this case the demands focus on the knowledge structures e. Often, the analysis is performed by assuming a computational model of the relevant cognitive processes, and the specific analysis approaches depend on the model adopted. Many techniques are used for the collection of task data, including observation, interviews, questionnaires, and review of instruction manuals.
The human factors literature is often used to find the range of capabilities in the appropriate population to compare with the task demands. A critical human factors method that is particularly appropriate for the design of components of health care systems is user testing. These tests may take the form of focus groups or usability testing with early mock-ups or mid-stage prototypes or final system components. Often in usability testing, a variety of prototypes or mock-ups are used.
For example, in the early stages of usability testing, two-dimensional representations of a device or user interface a graphical, nonfunctioning version of a system or storyboards that describe in a series of images the steps involved in execution of a task may be used, whereas working prototype devices or fully interactive systems may be used in later stages of testing. Frequently, especially in software engineering, human factors specialists use iterative prototyping, involving a series of tests with rough prototypes and short revision cycles National Research Council, In usability testing it is important to ensure that the participants are representative of the anticipated user groups and that the data collection techniques capture both the demands associated with the activities they will be performing and the relevant environmental contexts.
This is especially important with respect to health care systems for the home, for which the potential user groups are broad and diverse. Usability metrics include measures of effectiveness e. Human factors specialists rely on a variety of sources of information to guide their involvement in the design process.
This may initially include review of the existing literature, data compendiums, and design standards and guidelines. Databases that contain information on human capabilities are also available e. The document provides human dimension data and explains proper techniques for applying these data, which may vary depending on the complexity of the population to be accommodated. The guide includes a long list of resources and references Human Factors and Ergonomics Society, In addition, a number of design standards and guidelines are available to guide the design process of medical devices and systems see Chapter 5.
Human factors methods and knowledge can be applied to any stage of design or implementation of a system. This includes the initial design of systems and system components to avoid problems and deficiencies, as well as the diagnosis and identification of problems with existing systems.
Thus, the concepts and methods of human factors have broad applicability to health care in the home. For example, human factors techniques can be applied to the design of health care equipment and technologies, such as medication dispensers, glucometers, nebulizers, blood pressure monitors, telemedicine technologies, and software interfaces for Internet health applications.
These techniques can also be applied to the design of instruction manuals and training programs to ensure that individuals or their caregivers have the information and skills they need to operate equipment and perform health care tasks. Human factors techniques can be used to inform the design of a home environment to ensure that lighting, layout, and space are adequate for the tasks being performed or the design of a neighborhood to help ensure that there is adequate and effective signage. Human factors approaches are also relevant to the design of jobs for health care workers.
For example, human factors methods can be used to determine workflow, to coordinate work, to maintain scheduling and communication protocols, and to determine work requirements to ensure worker productivity, safety, and health. Human factors can have input into the broader organizational environment to help design and implement safety programs, certification protocols, or program evaluation methods.
Human factors techniques can also be used to help understand the sources of human errors and safety violations in the health care domain. In fact, the goals of human factors are commensurate with the goals stated in the report Crossing the Quality Chasm Institute of Medicine, for health system reform: safety, effectiveness, patient-centeredness, timeliness, efficiency, and equality. There are numerous examples in the health care domain in which the application of human factors has resulted in reduced errors and cost, increased safety, efficiency, and effectiveness, and personal satisfaction.
These examples include efforts to enhance safety and reduce medical errors e. Human factors methods have also been applied to the design of medical equipment and devices e. This is in contrast to a more traditional reductionist approach, which focuses on one component of a system in isolation from the other components.
Using a traditional approach, the focus is typically on the physical or technical components of a system, with little regard for the human. For example, glucometer or medication instructions may be designed without considering how the persons using these instructions might vary in terms of age, cognitive and sensory capabilities, English literacy, health literacy, or stage of illness acceptance.
For example, such errors could include lapses in performing health promotion and disease prevention behaviors, not adhering to a prescribed treatment, ignoring warning signs of complications, and not sharing important information about health history, symptoms, or response to treatment with caregivers. Other examples include potentially life-threatening events, such as misreading output from health monitoring equipment, altering equipment settings, turning off alarms, sustaining injuries due to poor body mechanics during lifting and transfers, or continuing intravenous IV antibiotic infusion in a person who is showing signs of allergic reaction.
There are many types of human error, and the causes and consequences of these errors vary. Although some errors may be inconsequential, others. Some errors and their consequences are preventable via good device or environmental design, whereas others must be handled through procedural or administrative solutions or through user education and training. They also require knowledge of whether the fit among these system elements is adequate. In summary, applying human factors knowledge and techniques to the design of health care systems intended for use in the home can make the systems safer, more effective, and more efficient.
Boff, K. Let denote the average NMI obtained for a fraction of randomly rewired links, where the links are chosen in random order,. By projecting the NMI between the exact- and reconstructed hierarchies, , to the axis using this function as. Based on that we define the linearized mutual information, LMI as. This quality measure corresponds to the fraction of unchanged links in a random link rewiring process, resulting in a hierarchy with the same NMI as.
By comparing the LMI to the fraction of exactly matching links, , we gain further information on the nature of the reconstructed DAG: If is significantly larger than , the reconstructed DAG is presumably better for the links high in the hierarchy, whereas if is significantly lower than , the reconstructed DAG is more precise for links close to the leafs.
Hierarchy in Natural and Social Sciences
Although the primary targets of tag hierarchy extraction methods are given by tagging systems with no pre-defined hierarchy between the tags, for testing the quality of the extracted hierarchy we need input data for which the exact hierarchy is also given. The corresponding input data for a tag hierarchy extraction algorithm would be a collection of proteins, each tagged by its function annotations.
Luckily, the Gene Ontology provides also a regularly updated large data set enlisting proteins and their known functions aggregated from a wide range of sources, a more detailed description of the data set we used is given in Materials and Methods. The matching between the two subgraphs is very good: the majority of the connections are either exactly the same shown in green , or acceptable shown in orange , by-passing levels in the hierarchy and e.
The appearing few unrelated— and missing links are colored red and gray, respectively. The exactly matching- and acceptable links are colored green and orange respectively, the unrelated links are shown in red, while the missing links are colored gray. The quality measures obtained for the complete reconstructed hierarchy are given in table 1. For comparison we also evaluated the same measures for algorithm B, the algorithm by P. Garcia-Molina, and the algorithm by P.
According to the results all 4 methods perform rather well, however, our algorithm seems to achieve the best scores. Although the ratio of exactly matching links is , which is not very high , the ratio of acceptable links is reaching , which is very promising. The corresponding plot showing the decay of the NMI between the Gene Ontology hierarchy and its randomized counterpart is given in Sect.
The large difference between and in favour of indicates that our algorithm is better at predicting links higher in the hierarchy.
Recommended for you
The reason why can stay relatively high for the reconstructed hierarchy is that the majority of the non-matching links are low in the hierarchy, therefore, have a smaller effect on the NMI. One of the most widely known tagging systems is given by Flickr, an online photo management and sharing application, where users can tag the uploaded photos with free words. Since the tags are not organized into a global hierarchy, this system provides an essential example for the application field of tag hierarchy extracting algorithms.
We have run our algorithm B on a relatively large, filtered sample of photos, the details of the construction of our data set are given in Methods. An example is given in Fig. The tags under these main categories seem to be correctly classified, e. More examples from our result on the Flickr data are given in Sect. Furthermore, similar samples from the hierarchies extracted by the other methods are also given in Sect. By running our algorithm B on a filtered sample from Flickr, we obtained a hierarchy between the tags appearing on the photos in the sample.
Stubs correspond to further direct descendants not shown in the figure, and the size of the nodes indicate the total number of descendants on a logarithmic scale, e. Another widely known online database is given by the IMDb, providing detailed information related to films, television programs and video games. One of the features relevant from the point of view of our research is that keywords related to the genre, content, subject, scenes, and basically any relevant feature of the movies are also available.
These can be treated similarly to the Flickr tags, i. The details of the construction of the data set are given in Methods. Although the tags appearing in the different sub-branches are all related to their parents, the quality of the Flickr hierarchy seemed a bit better. This may be due to the fact that keywords can pertain to any part of the movies, and hence, the tags on a single movie can already be very diverse, providing a more difficult input data set for tag hierarchy extraction. Nevertheless, this result reassures our statement related to the Flickr data, namely that the hierarchies obtained from our algorithm have a meaningful overall impression.
Similar samples from the hierarchies obtained with the other methods are shown in Sect. The results were obtained by running Algorithm B on a filtered sample of films from IMDb, tagged by keywords describing the content of the movies. Providing adjustable benchmarks is very important when testing and comparing algorithms. The basic idea of a benchmark in general is given by a system, where the ground truth about the object of search is also known.
However, for most real systems this sort of information is not available, therefore, synthetic benchmarks are constructed. Since the ground truth communities are known only for a couple of small networks, the testing is usually carried out on the LFR benchmark  , which is a purely synthetic, computer generated benchmark: the communities are pre-defined, and the links building up the network are generated at random, with linking probabilities taking into account the community structure.
The drawback of such synthetic test data is its artificial nature, however, the benefit on the other side is the freedom of the choice of the parameters, enabling the variance of the test conditions on a much larger scale compared to real systems. Here we propose a similar synthetic benchmark system for testing tag hierarchy extraction algorithms. The tag hierarchy extraction methods to be tested can be run on these sets of tags, and the obtained hierarchies, the "reconstructed" hierarchies , can be compared to the exact hierarchy used when generating the synthetic data.
When drawing an analogy between this system and the LFR benchmark, our pre-defined hierarchy is corresponding to the pre-defined community structure in the LFR benchmark, while the generated collections of tags are corresponding to the random networks generated according to the communities.
To make the above idea of a synthetic tagging system work in practice, we have to specify the method for generating the random collections of tags based on the given pre-defined hierarchy. In general, the basic idea is that tags more closely related to each other according to the hierarchy should appear together with a larger probability compared to unrelated tags.
To implement this, we have chosen a random walk approach as suggested in . The first tag in each collection is chosen at random. For the rest of the tags in the same collection, with probability we start a short undirected random walk on the hierarchy starting from the first tag, and choose the endpoint of the random walk, or with probability we again choose at random. An illustration of this process is given in Fig. The parameters of the benchmark are the following: the pre-defined hierarchy between the tags, the frequency of the tags when choosing at random, the probability for generating the second and further tags by random walk, the length of the random walks, the number of objects and finally, the distribution of the number of tags per object.
Although this is a long list of parameters, the quality of the reconstructed hierarchy is not equally sensitive to all of them. The objects in this approach are represented simply by collections of tags. For a given collection, the first tag is picked at random, illustrated in red , while the rest of the tags are obtained by implementing a short undirected random walk on the DAG, starting from the first tag, illustrated in purple. In Table 2. In the data generation process the exact hierarchy was set to a binary tree of 1, tags, with tag frequencies decreasing linearly as a function of the depth in the hierarchy.
We generated an average number of 3 co-occurring tags on altogether 2,, hypothetical objects, with random walk probability of and random walk lengths chosen from a uniform distribution between 1 and 3. We ran the same algorithms on the obtained data as in case of the protein data set, and used the same measures for evaluating the quality of the results. According to Table 2. Interestingly, the results of the algorithm by Schmitz were very poor on this input. Nevertheless, this method is still competitive with the others, e.
However, the study of why does this algorithm behave completely different from the others on our benchmark is out of the scope of the present work. In Table 3. According to the studied quality measures, the performance of the involved methods drops down drastically compared to Table 2. Garcia-Molina it is reduced to. This shows that algorithm B can have a significantly better performance compared to other algorithms, as the quality of its output is less dependent on the correlation between tag frequencies and level depth in the hierarchy.
Another interesting effect in Table 2. As we mentioned earlier, studies of the reasons for the outlying behavior of this algorithm on our benchmark compared to the other methods is left for future work.
The effects of the modifications in the other parameters of the benchmark are discussed in Sect. Nevertheless these results already show that the provided framework can serve as versatile test tool for tag hierarchy extraction methods. Both algorithms introduced in the paper depend on the -score related to the number of co-occurrences between a pair of tags. If the tags are assigned to the objects completely at random, the distribution of the number of co-occurrences for a given pair of tags and follows the hypergeometric distribution: Assuming that tag and appear altogether on and objects respectively, let us consider the random assignment of tag among a total number of objects.
Based on this, the probability for observing a given number of co-occurrences between and is. The -score is defined as the difference between the observed number of co-occurrences in the data, , and the expected number of co-occurrences at random as given in 5 , scaled by the standard deviation according to 6 ,.
For discrete variables and with a joint probability distribution given by , the mutual information is defined as. If the two variables are independent, , thus, becomes. The above quantity is very closely related to the entropy of the random variables,. Based on 9 , the NMI can be defined as. This way the NMI is 1 if and only and are identical, and 0 if they are independent. Both the exact DAG describing the hierarchy between protein functions and the corresponding input data set given by proteins tagged with known function annotations were taken from the Gene Ontology .
We concentrated on molecular functions, where the complete DAG has altogether 6, tags. However, a considerable part of these annotations are rather rare, thus, reconstructing the complete hierarchy would be a very hard task due to the lack of information. For the data set of proteins, tagged with their known molecular function annotations, we took the monthly quality controlled release as in The resulting smaller data set contained 5,, proteins, each having on average 3.
Flickr provides the possibility for searching photos by tags, thus, as a first step we downloaded photos resulting from search queries over a list of 68, English nouns, yielding altogether 2,, photos, the same photo can appear multiple times as a result for the different queries. At this stage we stored all the tags of the photos and the anonymous user id of the photo owners as well.
Next, the set of tags on the photos had to be cleaned: only English nouns were accepted, and in case of parts of a compound word appeared beside the compound word on the same photo, the smaller parts were deleted, leaving only the complete compound word.
Hierarchy in Natural and Social Sciences (Methodos Series, vol.3) - PDF Free Download
Since our algorithms rely on the weighted network of co-appearances, we applied a further filtering: a link was accepted only if the corresponding tags co-appeared on photos belonging to at least 10 different users. The resulting tag co-appearance network had 25, nodes, encoding information originating from 1,, photos. The goal of the keywords is helping the users in searching amongst the movies, and keywords can pertain to any part, scene, subject, gender, etc.
Although keywords can be given only by registered users, there is no restriction what so ever for registering, and the submitted information is processed by the "Database Content Team" of the IMDb site. The version of the original data we are used here contained , movie titles and , different keywords. However, to improve the quality of the data set, we restricted our studies to keywords appearing on at least a different movies, leaving , movies and 6, different keywords in the data set.
We introduced a detailed framework for tag hierarchy extraction in tagging systems. First, we have defined quality measures for hierarchy extraction methods based on comparing the obtained results to a pre-defined exact hierarchy. A part of these quantities were simply given by fractions of links fulfilling some criteria, e. However we also defined the NMI between the exact- and the reconstructed hierarchies, providing a quality measure which is sensitive also to the position of the non-matching links in the hierarchy. This was illustrated by our experiments comparing a hierarchy to its randomized counterpart, where the NMI showed a significantly faster decay when the rewiring was started at the top of the hierarchy, compared to the opposite case of starting from the leafs.
Furthermore, we developed a synthetic, computer generated benchmark system for tag hierarchy extraction. This tool provides versatile possibilities for testing hierarchy extraction algorithms under controlled conditions. The basic idea of our benchmark is generating collections of tags associated to virtual objects based on a pre-defined hierarchy between the tags. By running a tag hierarchy extraction algorithm on the generated synthetic data, the obtained result can be compared to the pre-defined exact hierarchy used in the data generation process.
According to our experiments on the benchmark, by changing the parameters during the generation of the synthetic data, we can enhance or decrease the difficulty of the tag hierarchy reconstruction. In addition, we developed two novel tag hierarchy extraction algorithms based on the network approach, and tested them both on real systems and computer generated benchmarks. In case of the tagged protein data the similarity between the obtained protein function hierarchy and the hierarchy given by the Gene Ontology was very encouraging, and the hierarchy between the English words obtained for the Flickr and IMDb data sets seemed also quite meaningful.
The computer generated benchmark system we have set up provides further possibilities for testing tag hierarchy extraction algorithms in general. By changing the parameters during the input generation we can enhance or decrease the difficulty of the tag hierarchy reconstruction.
Our methods were compared to current state of the art tag hierarchy extraction algorithms by P. Garcia-Molina and by P. Interestingly, the rank of the algorithms according to the introduced quality measures was varying from system to system. In case of the protein data set algorithm A was slightly ahead of the others, while the rest of the methods achieved more or less the same quality. In turn, for the easy synthetic test data, algorithm B and the algorithm by P. Garcia-Molina reached almost perfect reconstruction, with algorithm A left slightly behind, and the algorithm by P.
Schmitz achieving very poor marks. However, when changing to the hard synthetic test data, a large difference was observed between the quality of the obtained results, as algorithm B significantly outperformed all other methods. The different ranking of the algorithms for the included examples indicates that tag hierarchy extraction is a non-trivial problem where a system can be challenging for one given approach and easy for another method and vice versa. Nevertheless the results obtained indicate that tag hierarchy extraction is a very promising direction for further research with a great potential for practical applications.