Context vector generation and retrieval

In a computer having a processor, a database, and a display unit, a computer implemented method of generating vectors representing information items for retrieval of the information items, the method comprising:

• selecting a set of R information items in the database;
• determining for the selected set of information elements an R×R mutual co-occurrence matrix based on proximal co-occurences of the information items in a plurality of documents;
• applying Singular Value Decomposition to the mutual co-occurrence matrix to produce a set of first context vectors;
• and
• wherein each first context vector is uniquely associated with one of the selected information items, and wherein information items having similar meaning have respective first context vectors with similar orientations in the vector space;
• and
• using said context vectors for retrieval of information items from said database;
• and
• displaying said retrieved information items on said display unit.

In a computer having a processor, a database, and a display unit, a computer implemented method of retrieving a record from the database containing a plurality of records, each record containing at least one information item having an associated context vector, the method comprising:

• for each of a plurality of information items, storing context vectors associated with the information item the context vectors having the properties that information items having similar meaning have context vectors with similar orientations in a vector space, and information items having dissimilar meanings have context vectors with dissimilar orientations in the vector space;
• for each of the plurality of records from said database containing said plurality of records, storing a summary context vector derived from context vectors respectively associated with information items that comprise the record;
• receiving a query;
• deriving at least one query information item from the query;
• generating a query context vector from the query information item;
• and
• selecting at least one record from said database having a summary context vector with orientation in the vector space that is similar to the orientation of the query context vector;
• and
• displaying said selected at least one record on said display unit.

In a computer having a processor, a data storage, and a display unit, a computer implemented method of providing a space for information items, said information items encoding and representing predetermined meanings, the method comprising:

• selecting a set of first information items in a corpus of records in said data storage;
• creating a first set of context vectors based on proximal co-occurrences of the first information items, each first context vector uniquely associated with one of the first information items, the context vectors having an orientation in a vector space, such that first information items having similar predetermined encoded meaning have context vectors with similar orientations in the vector space;
• selecting a set of second information items in said data storage, the second information items being different from the first information items;
• selecting a subset of the first information items;
• for each first information item in the subset, selecting a corresponding second information item representing a predetermined encoded meaning substantially identical to the predetermined encoded meaning of the first information item;
• for each of the selected second information items, associating the second information item with the context vector of the corresponding first information item;
• assigning a context vector to each non-selected second information item;
• adjusting the context vectors of the non-selected second information items using the context vectors of the selected second information items;
• and
• using any of said context vectors for later retrieval of said first information items;
• and
• displaying said retrieved first information items on said display unit.

In a computer having a processor, a database of records, and a display unit, a computer implemented method of generating a dictionary of information items for said database of records, each record including at least one information item, each information item associated with a context vector, each information item having a determinate proximity to other information items in a record, wherein a neighbor information item is an information item that occurs proximate a target information item in at least one record in the database, the method comprising:

• initializing the context vectors such that initial context vectors are substantially orthogonal to each other in a vector space;
• associating the context vectors with information items in the dictionary of information items for said database of records;
• and
• for each information item being a target information item:
• selecting neighbor information items of the target information item in at least one record;
• and
• modifying the context vector of the target information item using the context vectors of each selected neighbor information items as a function of the proximity of each neighbor information item to the target information item, and a co-importance of the target information item and the neighbor information item;
• using said dictionary for retrieving information items representing media during a search;
• and
• displaying said retrieved information items on said display unit.

In a computer system including a storage device containing a plurality of records, each record containing a plurality of information items, a computer readable medium for configuring and controlling the computer system to generate a plurality of context vectors, the computer readable medium comprising:

• an initial context vector generation module, adapted to read and write to the storage device, which initializes to each of a plurality of selected information items an initial context vector, such that the initial context vectors are substantially orthogonal to each other in a vector space, and which writes the initial context vectors to the storage device in association with respective information items;
• a vector training module, adapted to read and write to the storage device, for modifying the context vector of a selected information item, being a target information item, using the context vectors of neighbor information items that proximally co-occur with the target information item, as a function of the proximity of each neighbor information item to the target information item, and a co-importance of the target information item and the neighbor information item;
• a retrieval module configured to use said initial context vector generation module and said vector training module in retrieving information items or records containing said information items, said information items representing media;
• and
• a display module for displaying said retrieved information items or records.

In a computer having a processor, a database, and a display unit, a computer implemented method of automatically indexing documents using a defined index of terms, the method comprising:

• providing an indexed collection of documents in said database, each document having at least one index term assigned to the document;
• providing a plurality terms, including the index terms, each term associated with a context vector, the context vector having the properties that that terms having similar meaning have context vectors with similar orientations in a vector space, terms having dissimilar meanings have context vectors with dissimilar orientations in the vector space, and terms which frequently proximally co-occur have context vectors with similar orientations in the vector space;
• and
• generating for each indexed document a context vector from the context vectors of selected terms that comprise the document;
• receiving a new document in said database to be indexed;
• generating a new context vector of the new document, the new context vector generated from the context vectors of selected terms that comprise the new document;
• selecting at least one indexed document having a context vector similar to the new context vector;
• assigning to the new document at least one index term assigned to a selected indexed document;
• such that said new document can later be retrieved from said database;
• and displaying said retrieved document on said display unit.