Dynamic Taxonomy process for browsing and retrieving information in large heterogeneous data bases

• US 7,778,993 B2
• Filed: 01/31/2008
• Issued: 08/17/2010
• Est. Priority Date: 12/16/1998
• Status: Expired due to Fees

• Current Assignee: Giovanni Sacco
• Original Assignee: 21 srl
• Inventors: Sacco, Giovanni
• Primary Examiner(s): Rones; Charles
• Assistant Examiner(s): Quader; Fazlul
• Application Number:
  US12/023,719
• Publication Number:
  US 20080133490A1
• Time in Patent Office: 929 Days
• Field of Search: 707/705
• US Class Current: 707/705
• CPC Class Codes: G06F 16/3323 using document space presentation or visualization, e.g. category, hierarchy or range presentation and selection
• CPC Class Codes: G06F 16/34 Browsing; Visualisation therefor
• CPC Class Codes: G06F 16/904 Browsing; Visualisation therefor for navigating the web G06F16/954; browsing optimisation for the web G06F16/957
• CPC Class Codes: Y10S 707/954 Relational
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99937 Sorting
• CPC Class Codes: Y10S 707/99942 Manipulating data structure, e.g. compression, compaction, compilation
• CPC Class Codes: Y10S 707/99944 Object-oriented database structure
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing

Abstract

A process is disclosed for retrieving information in large heterogeneous data bases, wherein information retrieval through visual querying/browsing is supported by dynamic taxonomies; the process comprises the steps of: initially showing (F1) a complete taxonomy for the retrieval; refining (F2) the retrieval through a selection of subsets of interest, where the refining is performed by selecting concepts in the taxonomy and combining them through boolean operations; showing (F3) a reduced taxonomy for the selected set; and further refining (F4) the retrieval through an iterative execution of the refining and showing steps.

Claims

1. A method for retrieving information from databases, said databases being structured or unstructured, said databases being homogeneous or heterogeneous, wherein retrieval is performed through visual queries on dynamic taxonomies, said dynamic taxonomies being an organization of concepts that ranges from a most general concept to a most specific concept, said concepts and their generalization or specialization relationships may be being called an intension, items in said databases being classified under one or more concepts, said items and their classification being called an extension, said method comprising:

using a computer for providing a taxonomy for said retrieval;

using the computer for operating on a selected subset of interest of said taxonomy in order to refine said retrieval, said selected subset of interest being specified by using the computer for combining selected taxonomy concepts through boolean operations or being specified through querying methods, said querying methods retrieving classified items according to different selection criteria;

providing a reduced taxonomy for said selected subset of interest, said reduced taxonomy being derived from said taxonomy by using the computer for eliminating from the extension of said taxonomy all items not in said selected subset of interest and by pruning concepts under which no item in said selected subset of interest is classified; and

using the computer for iteratively repeating said steps of operating on a selected subset of interest and of providing a reduced taxonomy to further refine said retrieval, wherein;

said step of pruning concepts includes eliminating from the taxonomy all the concepts under which no item in the selected subset of interest is classified, or preventing said concepts from being selected in order to specify interest sets;

said step of providing a reduced taxonomy either reports only the concepts belonging to the reduced taxonomy or, for each such concept also reports how many items in the interest set are classified under the concept;

said intension is organized as a hierarchy of concepts or as a directed acyclic graph of concepts, thereby allowing a concept to have multiple fathers;

items in said classification are classified programmatically or automatically;

in said extension, there exists at least one item such that said item is classified under at least two different concepts such that each of said two concepts is neither an ancestor nor a descendant of the other concept in the intension; and

said method to reconstruct all the relationships between any two concepts based on the classification by using the computer, a relationship between any two concepts existing if at least one item is classified (1) under a first concept or any descendants of the first concept, and (2) under a second concept, or any descendants of the second concept.

2. The method of claim 1, wherein each item is identified by a unique identifier, and, for each concept, there exists a deep classification that lists all items classified under said concept or any descendants of said concept in the taxonomy, and an optional shallow classification that lists all items directly classified under said concept.

3. The method of claim 2, wherein boolean operations on concepts are implemented through corresponding set operations on the deep classification of said concepts.

4. The method of claim 2, wherein said step of providing a reduced taxonomy for the selected subset of interest comprises a testing operation such that a concept is provided in the reduced taxonomy if an intersection between the selected subset of interest and the deep classification of the concept is not empty, the testing operation being configured to optionally count a number of items in said intersection to show a user a number of items in the selected subset of interest that are also classified under the concept, said testing operation being also configured to be applied to the shallow classification, if used, to show the user a number of items in the selected subset of interest that are also directly classified under the concept, the number being useful when items are classified at any level of abstraction in the taxonomy, said testing operation being also configured to be applied to a set including a single item, in order to compute a classification of said item, if not explicitly stored, said testing operation being also used to produce a reduced tree by testing and providing sons of a root and, on subsequent explosion of a concept, testing and providing sons of said concept.

5. The method of claim 2, wherein deep classifications are physically stored as uncompressed or compressed bit vectors, and a counting of items in a result of logic operations on bit vectors is performed through a constant table whose size is 2n, whose i-th element contains a number of bits at 1 in binary number i, and processing the uncompressed form of the bit vector n bits at a time, adding to a counter, for every group j of n bits, contents of the element of said constant table whose index in said table is equal to the binary value of said group j of n bits.

6. The method of claim 1, wherein said classification is implicitly stored as virtual concepts in external databases, said virtual concepts being concepts for which neither actual sons, that are actual values of a domain to be represented, nor an actual classification are stored, but instead are computed, said virtual concept being a simple virtual concept, which is described by four abstract operations:

given a virtual concept, retrieve all its sons;

given a virtual concept, retrieve deep classification;

given a son of a virtual concept, retrieve deep classification; and

given an item, find all the terminal concepts, descendants of the virtual concept, under which the said item is classified;

said virtual concept is a derived virtual concept, which is a simple virtual concept with additional restrictions.

7. The method of claim 1, wherein said dynamic taxonomy is used to represent data represented by a single view on an external database, said items corresponding to tuples, rows, records, or objects in said view and, in order to identify an item, a candidate key of said view being used as item identifier or two abstract relations being kept for mapping system-generated item identifiers to and from a primary key of the view, a taxonomy for said view being able to be constructed by inserting concepts of interest for said view in the taxonomy, each concept being associated to a boolean clause, said boolean clause being able to reference any attribute of a tuple in said view and returning true if and only if said tuple must be classified under said concept, said concept being a real concept or a virtual concept, said virtual concept using said view for a synthesis of sons and extensions.

8. The method of claim 4, wherein the reduced taxonomy is computed in a single step by applying said testing operation, for each concept in the intension, and further in that said operation is made more efficient through the steps of:

numbering the concepts in the intension;

initializing a table, the i-th element of which holds the status of the concept with number i, initialized at a pending status;

starting from uppermost levels of the taxonomy, and continuing down in the taxonomy, processing for each concept;

if the element of said table corresponding to said concept is empty, determining that said concept does not belong to said reduced taxonomy, and continuing the processing with a next concept;

if the element of said table corresponding to said concept is empty, applying said testing operation to the concept;

if said testing operation produces a non-empty intersection, determining that i belongs to said reduced taxonomy;

otherwise, determining that neither said concept nor any of the descendants belong to said reduced taxonomy and setting to empty all the elements in said table, corresponding to all the descendants of said concept in the taxonomy, said descendants being either computed from the intension or being precomputed and stored in a table which holds for each concept in the taxonomy a list of all concepts descending in the taxonomy, such a table being recomputed every time the intension changes.

9. The method of claim 1, said method accounting for the popularity of items by representing a measure of popularity through a branch in the taxonomy, wherein said measure of popularity for an item is derived from the number of predefined actions, such as access or purchase, applied to said item, or is supplied by an external source, such as user or editorial reviews.

10. The method of claim 1, said method allowing the personalization of interaction by predefining a focus, or a selected subset of items, and having the system use said focus instead of the universe for all user interactions.

11. The method of claim 1, said method allowing the restriction of access for a specific user to a predefined set of items, where different users are restricted to different sets of items, and for all the interactions of said specific user said set of items is used as a focus instead of the universe.

12. The method of claim 1, said method allowing tracking of user interests by keeping track of all the concepts used to define each focus, or selected subset of items.

13. The method of claim 1, said method performing the computation of the reduced taxonomy for a set of items, said method working on a relation containing a pair of attributes, one of said attributes representing item identifiers, the other one of said attributes representing concept identifiers in such a way that if there is in such relation a tuple with a specific item identifier and a specific concept identifier then the item identified by said specific item identifier is classified under the concept identified by said specific concept identifier, said method comprising retrieving all the tuples containing an item identifier belonging to said set of items;

counting the number of tuples in said retrieved tuples for each concept identifier, and producing said reduced taxonomy, concepts in said reduced taxonomy being all the concept identifiers with a count larger than zero, and the ancestors of said concepts.

14. The method of claim 1, further comprising saving computed interest sets in order to avoid recomputation.

15. The method of claim 1, said method further comprising saving computed reduced taxonomies in order to avoid recomputation.

16. The method of claim 1, said method applied to the retrieval of real estate items, wherein:

said intension also comprises a set of features, each of said features being a descendant concept of the root concept of said organization, each of said features having as descendants in the taxonomy a set of concepts, each concept in said set of concepts representing either a single value or a set of values for said feature;

said real estate items are classified, for each said feature, under zero or more concepts representing either a single value or a set of values for that feature.

17. The method of claim 1, said method applied to the retrieval of person records for applications such as matchmaking, social networks or human resource management wherein:

said intension also comprises a set of features, each of said features being a descendant concept of the root concept of said organization, each of said features having as descendants in the taxonomy a set of concepts, each concept in said set of concepts representing either a single value or a set of values for said feature;

said person records are classified, for each said feature, under zero or more concepts representing either a single value or a set of values for that feature.

18. The method of claim 1, said method applied to diagnostic applications such as, for example, medical diagnosis or malfunction diagnosis wherein:

said intension also comprises a set of features comprising symptoms, each of said features being a descendant concept of the root concept of said organization, each of said features having as descendants in the taxonomy a set of concepts, each concept in said set of concepts representing either a single value or a set of values for said feature;

said items are classified, for each said feature, under zero or more concepts representing either a single value or a set of values for that feature.

19. The method of claim 1, said method further comprising a geographic map, such geographic map being used to show the items in the selected subset of interest.

20. The method of claim 1, said method further comprising a geographic map, the area of such geographic map being modifiable by users, such area being used to refine the interest set by discarding from the interest set all the items whose geographic coordinates do not fall within such area.

23. The method of claim 1 or claim 22, said method applied to event log analysis wherein:

said intension comprises a set of features describing events, each of said features being a descendant concept of the root concept of said organization, each of said features having as descendants in the taxonomy a set of concepts, each concept in said set of concepts representing either a single value or a set of values for said feature;

said items are classified, for each said feature, under zero or more concepts representing either a single value or a set of values for that feature.

24. The method of claim 1 or claim 22, said method applied to social networks wherein:

said intension comprises a set of features describing an entity in the social network, each of said features being a descendant concept of the root concept of said organization, each of said features having as descendants in the taxonomy a set of concepts, each concept in said set of concepts representing either a single value or a set of values for said feature;

said entities are classified, for each said feature, under zero or more concepts representing either a single value or a set of values for that feature.

25. The method of claim 1 or claim 22, wherein items in said extension also comprise numerical attributes, and for each concept in the taxonomy, numerical attributes are presented as numerical values computed by an aggregating function from the values of the numerical attribute for all the items in the interest set that are classified under said concept, said aggregating function being predefined or defined.

21. A method for retrieving items from electronic catalogs, for applications such as electronic commerce or electronic auctions, wherein retrieval is performed through visual queries on dynamic taxonomies, said dynamic taxonomies being an organization of concepts that ranges from a most general concept to a most specific concept, said concepts and their generalization or specialization relationships being called an intension, said concepts also comprising features such as price, items in said electronic catalogs being able to be classified under one or more concepts, said items and their classification being called an extension, said method comprising:

using a computer for providing a taxonomy for said retrieval;

using the computer for operating on a selected subset of interest of said taxonomy in order to refine said retrieval, said selected subset of interest being specified by using the computer for combining selected taxonomy concepts through boolean operations, or being specified through querying methods, said querying methods retrieving classified items according to different selection criteria;

providing a reduced taxonomy for said selected subset of interest, said reduced taxonomy being derived from said taxonomy by using the computer for eliminating from the extension of said taxonomy all items not in said selected subset of interest and pruning concepts under which no item in said selected subset of interest is classified; and

using the computer for iteratively repeating said steps of operating on a selected subset of interest and of providing a reduced taxonomy to further refine said retrieval, wherein;

said hierarchical organization of concepts for said electronic catalogs comprises a set of features, each of said features being a descendant concept of the root concept of said organization, each of said features having as descendants in the taxonomy a set of concepts, each concept in said set of concepts representing either a single value or a set of values for said feature;

said items in said electronic catalogs are classified, for each said feature, under zero or more concepts representing either a single value or a set of values for that feature;

said step of providing a reduced taxonomy either reports only the concepts belonging to the reduced taxonomy or, for each such concept also reports how many items in the interest set are classified under the concept;

in said extension, there exists at least one item such that said item is classified under at least two different concepts such that each of said two concepts is neither an ancestor nor a descendant of the other concept in the intension; and

said step of pruning of concepts includes eliminating from the taxonomy the concepts under which no item in the selected subset of interest is classified, or preventing such concepts from being selected in order to specify interest sets.

22. A method for using a computer for retrieving association rules from databases, said databases being structured or unstructured, said databases being homogeneous or heterogeneous wherein retrieval is performed through visual queries on dynamic taxonomies, said dynamic taxonomies being an organization of concepts that ranges from a most general concept to a most specific concept, said concepts and their generalization or specialization relationships being called an intension, items in said databases being classified under one or more concepts, said items and their classification being called an extension, an association rule defining a probabilistic correlation relationship between the antecedent, said antecedent being defined by one concept in the taxonomy or by a boolean combination of concepts in the taxonomy, and the consequent, said consequent being defined by one concept in the taxonomy or by a boolean combination of concepts in the taxonomy, said method comprising:

using a computer for providing a taxonomy for said retrieval;

using the computer for operating on a selected subset of interest of said taxonomy in order to refine said retrieval, said selected subset of interest being specified by using the computer for selecting taxonomy concepts and combining taxonomy concepts through boolean operations, said selected concepts combined through boolean operations may be called a conceptual focus;

providing a reduced taxonomy for said selected subset of interest, said reduced taxonomy being derived from said taxonomy by using the computer for eliminating from the extension of said taxonomy all items not in said selected subset of interest and pruning concepts under which no item in said selected subset of interest is classified; and

iteratively repeating said steps of operating on a selected subset of interest and of providing a reduced taxonomy to further refine said retrieval, wherein;

said step of pruning of concepts includes eliminating from the taxonomy the concepts under which no item in the selected subset of interest is classified, or preventing said concepts from being selected in order to specify interest sets;

said intension is organized as a hierarchy of concepts or as a directed acyclic graph of concepts, thereby allowing a concept to have multiple fathers;

for each concept in said reduced taxonomy, two association rules exist, the first association rule having the conceptual focus of said reduced taxonomy as the antecedent of said first association rule and having said concept in the reduced taxonomy as the consequent of said first association rule, the second association rule having said concept in the reduced taxonomy as the antecedent of said second association rule and having the conceptual focus of said reduced taxonomy as the consequent of said second association rule;

in said step of providing a reduced taxonomy, for an association rule in the reduced taxonomy, a measure of confidence is provided, said measure of confidence being computed as the ratio between the number of items in the intersection of the antecedent and consequent of said association rule over the number of items in the antecedent of said association rule, or said measure is not provided;

in said step of providing a reduced taxonomy, for an association rule in the reduced taxonomy, a measure of support is provided, said support being expressed as the number of items in the intersection of the antecedent and consequent of said association rule over the total number of items, or said measure is not provided;

in said extension, there exists at least one item such that said item is classified under at least two different concepts such that each of said two concepts is neither an ancestor nor a descendant of the other concept in the intension; and

in said step of providing a reduced taxonomy, for an association rule in the reduced taxonomy, a measure of the statistical significance of how the subordinate probability of the consequent of said association rule with respect to the antecedent of said association rule deviates from independence of said consequent and antecedent of said association rule, is provided or said measure is not provided.

26. A method for the statistical comparison of different subsets of a database, said database being described by a dynamic taxonomy, said dynamic taxonomies being an organization of concepts that ranges from a most general concept to a most specific concept, said concepts and their generalization or specialization relationships being called an intension, items in said databases being classified under one or more concepts, said items and their classification being called an extension, said method comprising:

using a computer for initially providing a view for each of said subsets, said view being a reduced taxonomy derived from the initial taxonomy by setting a specific focus, or selected subset of interest;

using a computer for providing for each concept in each view, a measure of statistical deviation from uniformity for the subset represented by said concept with respect to the same concept in the first view only or in each of the other views including the first view, said measure of deviation only being a raw measure of deviation or including additional measures such as the statistical significance of such deviation, said first view being used as a reference view;

using a computer for selecting a subset of interest in any of said views, such selected subset of interest being automatically added to the selected subset of interest of each of the other views; and

repeating said steps of selecting subsets of interest and showing views.