KR100522839B1 - Incremental construction of thesauri,selective-spreading based browsing and intelligent query processing in electronic commerce - Google Patents

Abstract

The present invention relates to an object-oriented thesaurus construction, browsing and query method that can intelligently build and manage product search in an electronic commerce using an object-oriented paradigm. This paradigm's inheritance mechanism allows the expert to build the thesaurus progressively and conveniently by providing a structural grasp of the relationships between the concepts expressed in the thesaurus. In particular, when a large thesaurus is built by different experts on different hosts, the information captured by this mechanism helps maintain the thesaurus's semantic consistency, and the expert must specify all the relevance of the concepts. Minimizing the burden can reduce the cost of building a thesaurus. The intelligent object-oriented thesaurus system also provides selective extended browsing and intelligent query processing based on object-oriented methods.

Description

Incremental construction of the sauri, selective-spreading based browsing and intelligent query processing in electronic commerce}

delete

delete

delete

delete

delete

The present invention belongs to the technical field of the domain knowledge construction method for supporting effective search for specific information such as interior, chemical sites, electronic shopping malls in the information system of the electronic commerce. In particular, it is a technique for developing a thesaurus system that supports the establishment of a term recommendation method, a conceptual browsing method, and an intelligent query method direct approach by applying an object-oriented paradigm to a thesaurus.

Recently, as the accuracy and reproducibility become important as well as the speed of information retrieval system, many researches have been conducted on a system for constructing and effectively using a thesaurus. In particular, the prior art for thesaurus is the International Standardization Organization (ISO) 2788, which is a technology and domain expert that analyzes the frequency of simultaneous appearance of concepts from documents related to the application domain and automatically constructs thesaurus of the network structure and uses them for query processing. And 5964 standard, thesaurus structure is built directly, and the thesaurus concept that user wants is extracted and extracted through intelligent inference. These systems basically provide thesaurus building and maintenance, concept searching, and interworking with search systems.

Representative systems for constructing the network structure thesaurus include "PhraseFinder", "RBT", and "SSOM", which generally represent the degree of association between concepts. Here, the degree of relevance is automatically assessed through statistical information on the frequency of simultaneous appearance of concepts from domain documents. In particular, the co-occurrence frequency is used for constructing the correlation matrix and determining the dependency between words. The correlation matrix and the word dependency are known to be useful in the extended vector space model and the probability model, respectively. However, this thesaurus, which is automatically constructed by domain statistics, cannot accurately specify semantic relationships between concepts by the degree of relevance alone, and inevitably creates meaningless relationships that do not match our intuition. In addition, it has a disadvantage in that it is very difficult for a user to grasp the semantic relation between these concepts and refer to a desired concept.

Representative systems for building a layered thesaurus include "IBM Thesaurus Administrator", "WordNet", "NCI Thesaurus System", "BEAT", "Hierarch", "MultiTes", "TCS" and "STRID". Thesaurus includes "NASA", "INSPEC" and "Roget Thesaurus". The semantic relationship and degree of relevance of the concepts involved in this thesaurus are specified directly by the expert. In addition, in order to consistently construct thesaurus of accurate semantic structures and to maintain concepts with fluid characteristics, the expert has difficulty in grasping all the complex semantic structures among the concepts. In particular, because a large thesaurus can be built by multiple experts at the same time, it must be able to maintain mutual semantic consistency in the construction process, and it must include the ability to easily access the constructed thesaurus concepts. However, these systems for building a layered thesaurus only support a very simple sequential approach with hyperlinks to explore the user's desired concept from the thesaurus.

Also, as mentioned earlier, most information search engines, such as Naver, Empas, and Simmani, use the thesaurus almost synonymously, unlike this method of intelligently analyzing the user's intentions using terms contained in the query. I'm doing it.

The present invention relates to a method for developing an intelligent object-oriented thesaurus system that can express domain knowledge more structurally and intelligently by applying an object-oriented paradigm to a thesaurus. The technical problem to be achieved in this system is to solve the problems of the existing thesaurus system described above.

First, the structural characteristics of the object-oriented paradigm are applied to the thesaurus to accurately represent the semantic structure of domain knowledge. In other words, the thesaurus concept was regarded as an object, and the semantic relationship between objects previously expressed in BT / NT was defined as a generalization or classifying relationship to form an object hierarchy. Thus, many of the domain concepts are classified as instance objects and they can be effectively managed by concept objects that include many instance objects. In addition, RT, which has been interpreted in various meanings, is defined by subdividing into an aggregation relationship, an association relationship, and a relationship defined by the user. Objects that are associated with an object / aggregation relationship are property objects that can be inherited by subordinate objects. Becomes

Second, by using the inheritance property of object-oriented paradigm as a thesaurus construction mechanism, it is possible to construct and maintain thesaurus with consistent semantic structure in the system term recommendation method. That is, the system structurally grasps relationship information between upper objects and inherits the relationship information of lower objects. This inheritance mechanism supports the system's term recommendation-based thesaurus construction, minimizing thesaurus construction costs and maintaining the thesaurus's semantic consistency when multiple thesauruses are deployed on different hosts.

Third, it is possible to browse the thesaurus according to various viewpoints of the user by using the conceptualization technique of object-oriented paradigm. By abstracting the concept of object-oriented thesaurus, instance membership and relationship information between objects, you can sequentially browse thesaurus in a form that suits your needs. Conceptual abstraction is a way of progressively browsing concrete objects from objects in the general sense. In particular, this feature allows one to sequentially browse only the concepts in the user's intended direction among many sub-concepts derived from one top-level concept.

Fourth, it allows users to directly access thesaurus based query concepts. In particular, this feature supports a thesaurus direct reference method that can compensate for the disadvantage of multi-step browsing a large, complex thesaurus to find the user's intended concepts. That is, when a user expresses a thesaurus reference query using the concepts associated to find the exact objects he or she intended, the system infers objects that are semantically identical to the query and automatically derives the search query from these objects. Or optionally configured. Therefore, a search query constructed in this way will ensure a high accuracy search.

Fifth, the query processing used in the process of searching for the desired search is intelligent. For example, the system interprets this query in response to a request to find products that are semantically identical to the DVD or CDRW. After recognizing the facts, it searches for CD ROMs and DVCRs.

Figure 1 is an example of an intelligent object-oriented thesaurus that includes an object hierarchy where the top-level object is a 'computer part'. Thesaurus concepts in this example consist of terms that are often used in documents related to electronic shopping malls. Here, a node means an object and a link represents a relationship between the objects. The following description details the implementation of the term recommendation scheme, conceptualized browsing and intelligent thesaurus reference query processing flow diagram of the system for this example in detail with the accompanying drawings.

The intelligent object-oriented thesaurus system of the present invention is largely composed of a construction module, a browsing module, and an intelligent thesaurus reference query processing module. The following describes the technical problem of the present invention described above with reference to the accompanying drawings in detail. 1 is a basic configuration diagram of a hardware system to which the present invention is applied. The functions of each component are as follows. The main memory 101 of the system is equipped with a thesaurus system. The central processing board 102 is a board for executing a system mounted in the main memory. The input / output device 103 inputs the thesaurus content and outputs the CRT screen, text, and PostScript file. At this time, the thesaurus information input by the domain expert is processed by the object-oriented thesaurus system 105 and stored in the object-oriented database 104. The Unix operating system controls the boards and devices, and messages and information exchanged between the boards and devices are performed through the system bus 106.

2 is a basic process flow diagram for an object-oriented thesaurus system. The user selects the command mode of the system at 201. That is, a is a mode for thesaurus construction 203 of the term recommendation scheme of the system, and b is a mode for thesaurus browsing 204 of the conceptualization scheme. c is a thesaurus reference query processing 205 mode for intelligently extracting the associated concepts of a specific meaning from the thesaurus. 206 browses the constructed thesaurus hierarchy and related information and respective conceptualized thesaurus information. Relevant reference objects and their thesaurus reference graphs processed as a result of the thesaurus reference query according to user intention are visualized at 208. This process may be repeated at 208.

Figure 3 is a flow chart of the build module when the top-level object is created in the thesaurus, and this process is illustrated in detail in Figure 2. The object hierarchy of the 'operating system', 'computer component companies' and 'home appliances' has already been different. If built by an expert, one top-level object 'computer part' is created according to the following process. First, when the domain expert creates a 'computer component' at 301, at 302, presents the top-level object of another object hierarchy as an association object. Among them, when 'Operating System' and 'Computer Component Company' are selected, the degree of association 0.7 predefined in the association relationship is automatically specified in the relationship between them. In the same way, if the aggregation objects 'LCD', 'semiconductor device', 'CASE' and 'home appliances' are sequentially selected among the objects presented in (304), 0.8 is automatically assigned at (305). The degree of relevance automatically assigned can be redefined by an expert. In this example, the relevance of <part-of, computer parts, appliances, 0.8> has been redefined to 0.85.

4 is a flow chart for constructing the sub-objects of the term recommendation method of the system using the inheritance mechanism, and FIG. 3 illustrates this process. First, a generalization relationship is established by inserting the subordinate concept 'VGA' into the top-level object 'computer part' selected in 401 through 402 (403). Here, the negligible degree of 0.9 of the generalization relationship is automatically specified. In this process, it is very difficult to identify semantically coherent relationships by identifying all the objects that have an integration relationship and 'VGA', which is an expert inserted. In this case, the inheritance mechanism of the object-oriented thesaurus is very useful. In other words, the candidate objects are easily identified by the system because the aggregation / association objects of 'VGA' are selected among sub-objects of concepts having an aggregation / association relationship with the 'computer component'. First, the expert may set the relationship at 404 to either implicit (no) or explicit (yes). In the case of implicit relationship establishment, 405 and 406 automatically establish the relationship using the aggregation / association objects identified by the inheritance mechanism and the degree of association.

Figure 4 illustrates the process of identifying association candidate objects of 'VGA' and the degree of their relevance using the inheritance mechanism in explicit relationship establishment (407). This identified relationship information is presented to the expert and the expert can select only appropriate ones of this information at 408 to establish an association relationship. The degree of relevance among the selected relationship information may be appropriately redefined if necessary. This process is applied in the same way to the aggregation relationship setting 409 and 410.

In thesaurus, instance objects are the most specific objects that change more frequently than concepts, and are often used for query evaluation. Therefore, specifying the relationship of the exact instances is very important. Classifying relationships in object-oriented thesaurus allow you to efficiently maintain the thesaurus in the domain where many objects are treated as instances. In other words, by applying inheritance mechanisms to class relationships, experts can establish consistent and accurate relationships at lower cost when specifying instance relationships individually. This process is illustrated in Figure 5.

In Figure 5, 'GeForce' is a sub-instance object of 'VGA' and 'secondary storage device' selected in (401), and is inserted at (402) to a degree of relevance of 0.93 and 0.92, respectively, to establish generalization relationship (403). At this point, their degree of relevance explicitly described by the expert takes precedence over the implicit relevance degree 0.9 previously assigned to the classifying relationship. Next, using the inheritance mechanism, 'GeForce' is used by <association-of, VGA, Peripheral Company, 0.75> by 'Trident', 'Cirus Logic' and 'Creative Lab' and the implicit inheritance relationship 'association-of' (405). /0.69 'is set. The aggregation relationship is also set up in the same way at 406. If it is necessary to explicitly specify the aggregation / association relationship, the degree of association with the aggregation object automatically identified by the inheritance mechanism is presented at (407). If the expert selects `` Trident '', instead of the implied inheritance degree 0.69, it automatically suggests the relative degree 0.75 of the relationship <association-of, VGA, peripheral company, 0.75> of the parent object and the user corrects this degree accordingly. Can be specified (408). The system subsequently uses the associations <association-of, computer component, operating system, 0.7> in the same way to make candidate instances of Win, Unix, Win XP, and Linux the operating system. To establish relationships. The aggregation relationship is also set up in the same way.

5 is a process flow diagram for browsing a thesaurus constructed in a conceptualized manner. This process allows for sequential browsing of objects from the very complex thesaurus according to the user's point of view. Since the semantic structure of the object-oriented thesaurus is expressed in the object-oriented paradigm, it supports the selection mode 501 for browsing between objects according to the following three conceptualized user perspectives.

The "object conceptualization" mode is a browsing method through object conceptualization, which gradually browses a specific object from a general meaning object. In particular, this feature allows the user to sequentially browse only the objects in the user's intended direction among many sub-objects derived from one top-level concept. From 502 of Figure 1, when the user selects the top-level object 'computer part' in the general sense, (503) the sub-objects 'VGA', 'Multimedia Kit', 'Auxiliary Storage' derived from this object. , And identify the 'sound card'. The sub-objects and their relationships, which are identified information, are browsed at 504. Also, 'GeForce' is gradually browsed by selecting 'VGA' among the browsed objects.

The "Instance Affiliation Conceptualization" mode is a way of browsing instances that belong to one or more concepts in common. In general, a concept has many instances, and an instance belongs to several concepts. At this time, browsing through this conceptualization of the organization enables the user to clearly determine the relationship of the instance. If the user selects 'VGA' and 'secondary storage' at the same time from 505 in the example of Figure 1, (506) identifies the instances 'GeForce' and 'Millenium' that are common to these two objects (504). ) Browse these identified objects and their relationships. This feature is especially suitable for application environments with a large number of instances.

The "conceptual relation" mode is a method of selectively browsing only objects having a relationship desired by a user from one object. From the example of Figure 1, the user selects 'GeForce' at 507 and executes the relationship view command at 508 to refer to the information of the 'earphone' among the objects that have an aggregation / association relationship with 'GeForce'. do. 509 identifies the aggregation / association objects of 'GeForce' and browses them at 504.

6 is a process flow diagram for a thesaurus reference query processing that can directly reference objects required by a user from an object-based thesaurus expressed in complex relationships. First, at 601 a user enters a thesaurus reference query for concepts that they want to refer to, and 602 parses the syntax of this query via Lax / Yacc. At this time, the thesaurus reference query consists of two partial queries, such as [MAIN_QUERY, RESTRICTION_QUERY], where MAIN_QUERY selects a specific range of the hierarchy that the user wants to refer to in a comprehensive form. RESTRICTION_QUERY will further narrow the scope of concepts identified by MAIN_QUERY. In other words, Is one of the concepts identified by MAIN_QUERY. And at the same time it is limited to concepts that have an aggregation / association relationship, Is at least one It is limited to the concepts that are related to the integration and association with. 603 analyzes the meaning of the thesaurus reference query to infer appropriate reference objects and their degree of association from the thesaurus and visualize a reference graph for them (604). In this case, the reference graph is a hierarchical graph composed of all the high and low concepts of the inference objects and their relations. The user can select an automatic or optional mode at 605 for constructing a search query using reference objects. 606 automatically constructs a search query using all the reference objects including the degree of relevance and the Boolean operator "or". In (607), the user constructs a search query by using the user's desired objects among the reference objects and the appropriate Boolean operators "or" and "and". The former query can be useful in cases where a high reproducibility is required while ensuring a high degree of accuracy, and the latter can be used when a very high accuracy rate is required. They are used as a way to integrate with information retrieval systems.

7 is a detailed process flow diagram of the thesaurus object reference and reference graph generation of 603. First, 701 is a query that generalizes the type of MAIN_QUERY. Or query materialized Classify as Each case is illustrated by the examples in Figure 6 and Figure 7.

If the query is to be embodied, all objects at 702 Evaluates the set of sub-objects LBS contains in common. That is, when the thesaurus reference query is [VGA∧ secondary memory, *], the common sub-object set LBS = {GeForce, Millennium} of 'VGA' and 'VGA' is evaluated. (703) computes the set GLBS of the most common meanings for all objects in the LBS, in the example GLBS = LBS. These objects have the same meaning as MAIN_QUERY in the thesaurus reference query. In 706, for each object of GLBS, the set M of all parent objects is inferred. If RESTRICTION_QUERY does not exist, at 707, all objects of M evaluate the maximum degree of association with GLBS objects through the object hierarchy to compute a set of fuzzy reference objects and generate the corresponding generalization relationships as reference graphs. . The shaded parts in Figure 6 show the reference objects and the reference graph.

For generalized query cases, all objects at 704 Evaluates the set of parent objects UBS that are commonly included. For example, in the thesaurus reference query [GeForce ∨ Bud, *], the common set of ancestor objects for each of 'GeForce' and 'Bud' is UBS = {VGA, computer part}. 705 deduces the set LUBS of the objects in the most specific sense for every object of the UBS, where LUBS = {VGA}. These objects have the same meaning as MAIN_QUERY in the thesaurus reference query. In 706, for each object of LUBS, a set M = {VGA, computer component, GeForce, Bird, Millennium} of all the parent objects is identified. Between the materialized GLBS objects and the generalized LUBS objects, GLBS operations and LUBS operations can be reapplied repeatedly.

If RESTRICTION_QUERY exists, perform 708 and 709 on the RESTRICTION_QUERY object set M deduced by MAIN_QUERY. (708) for i = 1, ..., m of the objects of M Compute an object set, S _i, that has an aggregation / association relationship with. They perform 706 again through the result of performing the union operation at 809. In Figure 8, when M = {Computer Component, VGA, GeForce, Bud, Millennium} and RESTRICTION_QUERY = 'Win XP' AND 'Trident' AND 'Earphones', 'Win XP', 'Trident' at (708) and (709) And the element S = {GeForce} of M having a co-aggregation association with the earphone. Next, at 706, all the upper and lower object sets M = {computer part, VGA, auxiliary storage device, GeForce} of 'Geforce' are re-identified. At this time, the reference object set is presented by evaluating the maximum degree of relevance of all the objects of the set S and M through the object hierarchy, and generates the reference graph as shown in Figure 8. Reference queries can also be used to build progressive thesaurus. That is, even when the thesaurus is constructed, the candidate query thesaurus terms to be added from related documents can be easily found by performing the extended search based on the reference query using the built-in thesaurus even for the DB that is the target of the expansion query using the thesaurus terms. The additional thesaurus terms are used again in the expansion query to allow for progressive thesaurus construction.

Thesaurus system is very important in that the thesaurus can be built and maintained more effectively, and the thesaurus can be easily used by the search system or users to improve the search performance. To this end, the present invention invented a method for developing an object-oriented thesaurus system. This system consists of the function of constructing a thesaurus, the sequential browsing of related concepts using the conceptualization technique, and the intelligent reference of desired concepts through the reference query. The object-oriented paradigm applied in the construction process enables to accurately represent the relationships between objects and to build a thesaurus consistent with the intuition of the expert. In particular, the information structurally grasped by this paradigm's inheritance mechanism is actively used to construct thesaurus by recommendation of the system, and maintains the thesaurus's semantic consistency when multiple experts build on different hosts. It can also reduce the cost of deployment. Browsing and query-based referencing enable users to search the intended concepts sequentially or directly to construct accurate search queries. In particular, this method of query construction can be usefully used for domains that require high recall rates. A typical application domain is a software reusable parts search system.

Figure 1: Basic Diagram of Hardware System

Figure 2: Processing Flow for an Intelligent Object-Oriented Thesaurus System

Figure 3: Process Flow for Constructing Top-Level Thesaurus Objects

Figure 4: Process flow for constructing a thesaurus sub-object with progressive build

Fig. 5: Process flow diagram for selective extended browsing

Figure 6: Flowchart for Intelligent Query Processing

Figure 7: Processing Flow for Intelligent Object-Oriented Thesaurus Terminology Reference and Retrieval

Claims (5)

The hierarchy of thesaurus categories using the concept of object-oriented paradigm on the graphical thesaurus user interface visualized in a tree form can be viewed by double clicking on the upper concept, and by double clicking on the lower concept. One step (figure 5 of related drawing) in which the instances can be viewed and the hierarchy is simplified by double clicking on the upper concept again; A second step of freely moving and searching the category hierarchy of the thesaurus through the first step; Thesaurus management method using the above two steps A step 1 in which graphically and numerically the relationship between the thesaurus categories and instances belonging to several trees conceptualized at the same level is simultaneously displayed; In the first step, select and move the category and instance in the other tree whose relation is set, and then move to the upper and lower categories or instances using conceptualization centering on the category and the instance, or to simplify or expand to the upper category again. Seeing second step; Thesaurus management method using the above two steps (relevant drawing 5) If a relevance is set between the top two categories in the generalization hierarchy, then each subcategory and instance also inherits this relevance, and each particular subcategory and instance also inherits this relevance. A first step of presenting and recommending candidate categories and instances in the hierarchy so that subcategories and instances may be associated with each other; A second step in which a related type or degree of inheritance is automatically inherited and set only by a builder selecting some of the candidate categories and instances presented (relevant drawing 4); Thesaurus management method employing these two steps in a visualized environment Find thesaurus terms that have a specific relationship with (or defined by) the other terms in the thesaurus that specify scope in the query among subcategory subcategories and instances of the term representing the scope within the query, and use them to After searching, step 1 suggesting appropriate thesaurus candidate terms from the results (relevant figure 4); Adding the thesaurus candidate terms to the thesaurus tree for selection and repeating step 1 to enable construction of a progressive thesaurus; Thesaurus management method applying the above two steps repeatedly A first step of retrieving one or two higher categories that can generalize them from a user query consisting of an OR operation between terms; A second step of retrieving one or two specific subcategories or instances that can represent them from a user query consisting of an AND operation between terms; A third step of relational inference based reference query combining the two steps to infer a conclusion meant by the query; Thesaurus management method using the above three steps (relevant drawings 6 and 7)