General information
In numerous walks of life there is a need to easily compare and classify objects, to determine whether they are part of a group or collection characteristic of certain predefined features. The need stems from the necessity to detect features, to eliminate certain events, prevent critical situations and identify objects. The need is mainly generated by information retrieval systems, monitoring systems, recommendation, identification, moderation and numerous other systems, mainly focusing on a certain compound object and its relations with models.
In such cases compound objects are processed in order to reach a certain predefined goal (differing from one system to another). The problem of similarity is connected with comparators used as a tool for comparison. It is used to determine the level to which two objects are similar. In addition, this element makes it possible to build complex decision structures in form of a network of comparators which may be somewhat analogous to hierarchical classifiers (though not always and not limited solely to this role). It facilitates step-by-step modelling of decision-making and introducing changes or extension of the process.
The main possible fields of application of comparators relate to Decision Support Systems as direct application, still divided into narrower fields, namely:
CBR (Case-based reasoning) – systems basing on historical knowledge collected (cases solved in the past), where new solutions are produced by searching for previously positively concluded cases. The basic stages of decision-making by means of CBR include:
(a) seeking a good adaptation of the (new) problem in question to recorded historical cases, (b) adaptation of the previous solutions to the current problem, (c) finding solution to the problem in question and recording the result.
Application of comparators facilitates building solutions in the scope of (a) and partially (c).
CEP (Complex event processing) – system to monitor events in real time, used in industry. These systems analyse data from various sources, compare them and detect models, tendencies and exceptions related to events analysed. Consequently, undesirable situations are detected before they occur. This facilitates taking appropriate decisions to avoid them. Comparators support detection of models and facilitate the maintenance of such systems (adding new models, etc.).
RS (Recommender systems) – one of the decision-making systems, able to anticipate user preferences as regards the selection of subsequent element from resources available. Solutions like these are very popular in e-commerce (online shops) and entertainment (suggested films). Systems of this class have a common goal: to select the best subset of elements for each user. The selection depends on user behaviour, previous actions in short- and long-term. Any data can turn out important when taking the final decision. There are many ways of doing it. For example, one may use compound object comparators. The basic assumption is to determine preference grouping, user or resource cluster. Then, multiple similarities are studied in terms of various features. The final result is aggregated.
MS (Moderation systems) – systems to moderate content provided by systems (e.g. website portals, forums, chats, etc.). The aim of these solutions is to detect certain defined models and to block their publication or provide the relevant person with relevant information. Comparator-based solutions do very well in this field. One can easily build a solution that is easy and comprehensible to the final user. DituelModerator is one of the examples of finished products.
RMS (Risk Management Systems) – risk is a very complicated phenomenon. The computer system established for the purposes of risk management processes compound objects in order to calculate the value of risk for various aspects. Objects include processes, information, activities and numerous other elements in which risks are present. The ability to compare these complex problems is one of the tools to effectively implement such solutions. Risk management system created for the needs of the fire brigade (ICRA – project financed entirely from the funds of the National Centre for Research and Development) is one of the examples. The system assesses the level of risk of rescue and fire-fighting action in progress. The project uses comparators to identify the possible risks during the action and to manage the module, which searches for similar fire-fighting action scenarios.
Multimedia databases – database systems able to store multimedia data, e.g. photographs, films, music. Apart from keeping records, databases like these process, seek and aggregate information about compound objects in question in a manner that is effective and acceptable for the user. The role of these systems is to provide tools to communicate with the user, for questions asked to most effectively reflect user expectations. This field is a direct extension of classical relational databases (though not only), where the notion of compound object was rather equated with its description in form of relation tuples described by certain attributes. In the case of multimedia databases metadata of objects were enriched with the same objects (closed structures) which have to be compared and classified by database engines on similar principles to data in standard relational databases.
Search engines – search engines for any kind of information. In the majority of cases working on noisy data. Queries are often imprecisely and ambiguously coded.
Systems based on these engines optimise accuracy of query results retrieved in order to deliver information most desired by the user. Compound objects may take on forms of various text documents indexed by search systems (e.g. html websites) as well as images, sounds, films, etc. Depending on the type, these objects are subject to analysis in order to determine additional description resulting from their structure or content. Histogram text analysis is an example. The basic task is to calculate the arrangement of words in a text (insignificant words excluded) and, consequently, to ascribe the text to specific groups (categories).
Image analysis systems – wherever models are recognised, it is justified to use systems based on comparators. This is the case for image analysis. Image detection requires identification and recognition of numerous features. In this regard, comparator networks can fulfil many different tasks. Starting from character recognition, known as the OCR, through vehicle identification on the basis of number plates as well as image classification.
All areas listed above share the need of analysing similarity. The examples fail to provide in detail the way similarity is analysed. Similarity can be defined in various ways in various fields, systems and methods.
However, each case may involve universal methodology based on comparators used to compare the compound object with a collection of reference objects. Examples mentioned above reveal that compound objects differ in terms of type, application and significance. Irrespective from the similarities, however, the suggested methodology ensures a common approach to each case. The resulting development of comparative techniques refines the analysis of objects in numerous fields of exploitation simultaneously.
In the case of interest in the topic of comparators or the need for such solutions in a given activity (IT systems, companies, institutions), please contact us by means of the form available (here). We produce complete systems tailored to individual client's needs, using techniques described above. We also integrate submodules dedicated to a specific functionality with existing external systems.
Publications on the topic
Too see the list of publications click (link).
