CN103440314A - Semantic retrieval method based on Ontology - Google Patents

Abstract

The invention discloses a semantic retrieval method based on Ontology. Firstly, an ontology base is built, and building of ontology rules is further finished; search keys are input by a user, the value of similarity is worked out through concept similarity under the support of ontology; then, the concept that the value of the similarity is larger than a threshold value is added to an original search key set according to the set threshold value to form a new concept set in an extending mode; searching is carried out on the ontology base through the new concept set serving as source input words; finally, searching results are fed back to the user. According to the semantic retrieval method based on the Ontology, under the support of an ontology reasoning technology, accurate searching of living examples is achieved through the application of attributes in the ontology, and accordingly information retrieval recall ratio and precision ratio are improved when being compared with those in the prior art. According to the semantic retrieval method based on the Ontology, the grammar level of simple keyword matching is improved to the semantic level capable of being understood by a computer, and therefore input keywords can be understood by the computer, and accordingly retrieval intelligentization is achieved.

Description

A kind of semantic retrieving method based on Ontology

Technical field

The invention belongs to the crossing domain of natural language processing and machine intelligence, be exactly by Ontology (body) technology, current information retrieval technique is brought up to the intelligible semantic retrieval mode of computing machine from the keyword match mode, this technology is applied to the travel information retrieval upper, has realized intelligentized tourism retrieval service.

Technical background

Along with the fast development of Internet and mobile communication technology, Web has become global information source, how to find quickly and accurately own required information from immense information resources, becomes puzzlement user's a difficult problem.The retrieval mode that traditional information retrieval offers the user is the retrieval mode that the keyword with user's input is mated, but in most situation, this simple keyword coupling is difficult to understand the real retrieval purpose of user, therefore causes the degree of accuracy of current this information retrieval mode not high.

Some external experts have carried out the operation of practicality aspect body, and for example the Ontoseek yellow page system, be the system of an information retrieval based on contents, this system integration product tree and online Yellow Page.It combines system and body content matching mechanisms with representation ability, and ontology library and data base dictionary are combined, the user is provided one and can then be converted into the vocabulary in domain body by the input natural language, realize the searching system of semantic level.From the Ontoseek yellow page system, can learn, concept vocabulary wherein and the relation between vocabulary are under no restraint, so between vocabulary, perhaps graph of a relation cuts little ice, the result that therefore can retrieve is not that the user is required.Current tourism Yellow Page service system, as " ctrip.com ", " way ox net ", in retrieval, the key word information of user's input has just been carried out to the simple match of words, can not understand the information of user's input semantically, also just can not retrieve well the information that the user really needs.Therefore, the information retrieval mode must be raised to indescribably based on knowledge understanding rank, semantic-based rank and information be organized and expressed based on the keyword matching stage from existing, thereby design a kind of information retrieval model that is appreciated that user semantic.

Summary of the invention

For the above-mentioned problems in the prior art, the present invention proposes a kind of semantic retrieving method based on Ontology, purpose is to realize the understanding of computing machine to user input content, realizes the retrieval of semantic level.

For achieving the above object, the technical solution used in the present invention is: at first, build ontology library, complete the foundation of ontology rule.The search key of user's input, under the support of body, calculate the size of similarity by concept similarity.Then according to the threshold value of setting, the value of similarity is added in original search key set higher than the concept of threshold value, expand to new concept set.Using new concept set as source, the input word is retrieved in ontology library.Finally the result retrieved is returned to the user.

A kind of semantic retrieving method based on Ontology comprises the following steps:

Step 1, complete the structure of ontology library by the ontology development instrument, by manual type, analyze concept or the core vocabulary of tour field.The ontology library built forms the data structure of tree type.

Step 2, according to the relation between field concept, utilize Jena rule syntax form, completes the foundation in ontology rule storehouse.

Step 3, the user inputs retrieval vocabulary or statement, and the participle search engine carries out word segmentation processing.

Step 4, carry out the semantic retrieval expansion according to the calculating of concept similarity, forms new concept set.

Step 5,, retrieved as prime word with new set, and, under the support of ontology rule, body is carried out to reasoning, retrieves information implicit in ontology library.

Step 6, the result retrieved is sorted by the similarity size.

Step 7, return to the user by result for retrieval.

Compared with prior art, the present invention has the following advantages:

(1) the present invention, under the support of ontology inference technology, has realized accurately searching of example by the application to attribute in body, and the recall ratio of information retrieval, precision ratio are all increased than prior art.

(2) the method for the invention is brought up to the understandable semantic hierarchies of computing machine by the grammatical levels of key word simple match, makes computing machine can understand the keyword of input, thereby has realized the intellectuality of retrieval.

The accompanying drawing explanation

Fig. 1 is method flow diagram involved in the present invention;

Fig. 2 is embodiment of the present invention public transport result for retrieval;

Fig. 3 is embodiment of the present invention hotel result for retrieval;

Fig. 4 is embodiment of the present invention sight spot result for retrieval.

Embodiment

Below in conjunction with drawings and Examples, the present invention will be further described.

The software environment of needs of the present invention: Windows operating system, Myeclipse developing instrument, WEB server: Tomcat, database: Mysql, Spring+Struts+Hibernate framework.

The method of the invention process flow diagram as shown in Figure 1, comprises the following steps:

Step 1, complete the structure of ontology library by ontology development instrument Prot é g é, by manual type, analyze concept or the core vocabulary of tour field.The ontology library built forms the data structure of tree type.

Step 2, for example, according to the relation between field concept: the relation between public transport and station, attribute that can be such with " process " or " arrival " is set up contact, utilizes Jena rule syntax form, completes the foundation in ontology rule storehouse.

Step 3, the user inputs retrieval request: the user inputs retrieval request by search interface, submits to the background process program.

User interface utilizes Jsp and Jquery technology to realize, clicks index button and triggers Onclick () event, by the javascript power function, the keyword of page input is passed to the backstage code.

Step 4, the semantic retrieval expansion:

(1) keyword of user's input is carried out to word segmentation processing, form initialized set, wherein by Open-Source Tools Lucene.

(2) expand in whole ontology library with the initialization set in the same or analogous field concept of phrase semantic, form new concept set.

(3) calculate the Ontological concept similarity.

The computing formula of concept distance is:

Distance(a,b)＝N[a,Ancestor(a,b)]+N[b,Ancestor(a,b)]

$N[b,\mathrm{Ancestor}(a,b)]=\underset{x\∈\mathrm{path}(a,\mathrm{Ancestor}(a,b))}{\mathrm{\Σ}}\mathrm{Dis}\tan \mathrm{ce}(x,\mathrm{Parent}\left(x\right))$

$\mathrm{Dis}\tan \mathrm{ce}(x,\mathrm{Parent}\left(x\right))=\frac{\mathrm{\α}}{\mathrm{weight}(x,\mathrm{Parent}\left(x\right))}-\mathrm{\α}$

In formula, Distance (a, b) be illustrated in the distance of concept node a and concept node b in tree, Ancestor (a, b) be illustrated in the most recent co mmon ancestor node of concept node a and concept node b in tree, path (a, b) mean the set that on node a and the b shortest path in the tree of domain body formation, all concept nodes form, Distance (x, Parent (x)) be the distance between child node and father node, weight (x, Parent (x)) is the weights on the limit between this node and its father node, and α is adjustable factors.

In known art body tree, under the prerequisite of the distance of any two concept nodes, the similarity of any two concepts is:

$\mathrm{Similar}(a,b)=\frac{\mathrm{\θ}}{\mathrm{Dis}\tan \mathrm{ce}(a,b)+\mathrm{\θ}}$

In formula, θ is for regulating parameter.

The value of denominator is less, and the value of similarity is just larger.When Distance (a, b)=0, denominator is got minimum value θ, and concept similarity is got maximal value 1, and concept node a is similar to concept node b; When Distance (a, b)=∞, concept similarity is got minimum value 0, and concept node a and concept node b almost do not have any relation.

(4) concept that similarity is greater than to predetermined threshold joins in initialized set, again expands to new term set.Threshold value is the value calculated in repeatedly user's input, Multiple through then out similarity.That sets is excessive, the concept that very difficult expansion makes new advances; That sets is too little, and it is too large that the concept that expansion is come in and user input the semantic gap of keyword.The threshold value of the present embodiment gets 0.7.

Step 5, ontology inference: by the SPARQL query language, on the basis of the ontology rule built, the ontology model of establishment and the inference machine (inference machine is the inference engine that utilizes the Jena kit to carry) of registration are bound, then the ontology model data are carried out to reasoning, the result retrieved not only comprises already present instance data, has also comprised instance data implicit in the ontological resource, and the result then retrieval obtained preserves.Below provide 3 ontology rules:

Rule1:(?x?hasStation?y),(?y?HasBus?z),(?z?isAccessible?w)→(?x?isAccessible?w)

That is: near x, have y station, y station that z bus process is arranged, the z bus can arrive w, can arrive w from x so.

Rule2:(?x?Part?of?y)，(?z?Part?of?y)→(?x?Nearby?z)

That is: x belongs to the y area, and z also belongs to the y area, and x and z are adjacent so.

Rule3:(?x?hasRoute?y)，(?z?hasRoute?y)，(?z?isAccessible?w)→(?x?isAccessible?w)

That is: there is the y bus at the x station, and also there is the y bus at the z station, and z can arrive the w destination, and x also can arrive the w destination so.

Step 6, result for retrieval sequence: by the calculating of concept similarity, similarity, the larger front that is arranged in, inquired about at first, and the result of inquiry is exactly sorted result.

Step 7, return to the user by result for retrieval.

Below provide an application example of the present invention.

Search method proposed by the invention is applied to the retrieval of Tourism Resources of Beijing, the data that experiment is retrieved are all from resources bank Notology_tour.owl, this resources bank is the manual sort, then by the reptile instrument, the tourism data that crawl from internet, the data in resources bank are the web datas by crawler capturing.For the ease of experimental result, calculate, 1000 of the resource quantity of each classified information data selection, be set as 100 for the related resource that will retrieve, then retrieved.For example: hotel information is totally 1000 resources, and near 100 of associated hotel Beijing University of Technology.Below according to several different retrieval situations, analyzed:

(1) information of retrieval bus: the essential information that for example will retrieve Beijing No. 52 buses, retrieval clicked in the input key word, in resources bank, exist in the situation of data instance, example in this keyword and ontology library is compared to inquiry, by its attribute of ontology library case retrieval and the property value thereof after coupling, the information obtained is returned to front page layout, present to the user.Result for retrieval as shown in Figure 2.

(2) near the hotel of retrieval sight spot: the hotel information of Beijing University of Technology's website periphery of take is example, for example, in the zone at place, Beijing University of Technology station, find the hotel's information which has can stay, directly retrieval be perhaps inquiry less than, but can be by inference rule:

Rule4:[Rule4:(?xhotellocated?z)(?ystationLocated?z)->(?xhotelLocated?y)

If that is: the x hotel is positioned near z, the y hotel also is positioned near z, and the x hotel is adjacent with the y hotel so.

Then the Rule4 rule is imported in inference engine, the result set of the reasoning of process inference engine and the definition of SPARQL query statement, the result that retrieval obtains is exactly the needed result of user.Result for retrieval is as shown in Figure 3:

(3) near the sight spot retrieval website: for example inquire about near sight spot, station, Tian An-men, select corresponding route.If can not be directly and the data in ontology library mated, the semantic extension algorithm that system can realize through the present invention be expanded keyword, forms a new concept set, to enrich the semantic information of retrieval.Such as adding the concepts such as station, Zhan, Wangfujing, Dongdan in set, more just can retrieve required result in conjunction with reasoning.Calling rule Rule5:

Rule5:(?xIsNearby?y)(?yhasAttraction?z)->(?xhasAttraction?z)

If that is: x and y are adjacent, tourist attractions z is arranged near y, tourist attractions z is arranged near x so.

Then the Rule5 rule is imported in inference engine, the result set of the reasoning of process inference engine and the definition of SPARQL query statement, the result that retrieval obtains is exactly needed result.Result for retrieval is as shown in Figure 4:

Result for retrieval to Fig. 2～4 is added up, and 52 82 of public transport related resources, retrieve 144 altogether.Near 85 of hotel's correlated measures Beijing University of Technology, retrieve 130 altogether.Near Tian An-men, the sight spot related resource is 93, retrieves altogether 155.Spend a holiday recall ratio and precision ratio that net and semantic retrieving method of the present invention retrieved of application travel with partners is as shown in table 1.

The result comparison that table 1 the present invention and travel with partners spend a holiday and net retrieval

As shown in Table 1, traditional tour site that semantic retrieving method of the present invention is all representative higher than ctrip.com on recall ratio and precision ratio.The deficiency of tradition tour site is the intervention that there is no inference function, such as: retrieve the information of certain sight spot periphery, this retrieval mode that key word is mated is irrealizable.Data from above chart show and can draw, under the support of ontology, accuracy rate and the recall ratio of result for retrieval all increase.

Claims (2)

1. the semantic retrieving method based on Ontology, is characterized in that information retrieval is brought up to semantic retrieval from traditional keyword retrieval, comprises the following steps:

Step 1, complete the structure of ontology library by the ontology development instrument, by manual type, analyze concept or the core vocabulary of tour field;

Step 2, according to the relation between field concept, utilize Jena rule syntax form, completes the foundation in ontology rule storehouse;

Step 3, the user inputs retrieval vocabulary or statement, and the participle search engine carries out word segmentation processing;

Step 4, carry out the semantic retrieval expansion according to the calculating of concept similarity, forms new concept set;

Step 5,, retrieved as prime word with new set, and, under the support of ontology rule, body is carried out to reasoning, retrieves information implicit in ontology library;

Step 6, the result retrieved is sorted by the similarity size;

Step 7, return to the user by result for retrieval.

2. a kind of semantic retrieving method based on Ontology according to claim 1 is characterized in that step 4 is carried out the method for semantic retrieval expansion further comprising the steps of:

(1) keyword of user's input is carried out to word segmentation processing, form initialized set;

(2) expand in whole ontology library with the initialization set in the same or analogous field concept of phrase semantic, form new concept set;

(3) calculate the Ontological concept similarity;

The computing formula of concept distance is:

Distance(a,b)＝N[a,Ancestor(a,b)]+N[b,Ancestor(a,b)]

$N[b,\mathrm{Ancestor}(a,b)]=\underset{x\∈\mathrm{path}(a,\mathrm{Ancestor}(a,b))}{\mathrm{\Σ}}\mathrm{Dis}\tan \mathrm{ce}(x,\mathrm{Parent}\left(x\right))$

$\mathrm{Dis}\tan \mathrm{ce}(x,\mathrm{Parent}\left(x\right))=\frac{\mathrm{\α}}{\mathrm{weight}(x,\mathrm{Parent}\left(x\right))}-\mathrm{\α}$

In formula, Distance (a, b) be illustrated in the distance of concept node a and concept node b in tree, Ancestor (a, b) be illustrated in the most recent co mmon ancestor node of concept node a and concept node b in tree, path (a, b) mean the set that on node a and the b shortest path in the tree of domain body formation, all concept nodes form, Distance (x, Parent (x)) be the distance between child node and father node, weight (x, Parent (x)) is the weights on the limit between this node and its father node, and α is adjustable factors;

In known art body tree, under the prerequisite of the distance of any two concept nodes, the similarity of any two concepts is:

$\mathrm{Similar}(a,b)=\frac{\mathrm{\θ}}{\mathrm{Dis}\tan \mathrm{ce}(a,b)+\mathrm{\θ}}$

In formula, θ is for regulating parameter;

Denominator is less, and similarity is larger; When Distance (a, b)=0, denominator is got minimum value θ, and concept similarity is got maximal value 1, and concept node a is similar to concept node b; When Distance (a, b)=∞, concept similarity is got minimum value 0, and concept node a and concept node b almost do not have any relation;

(4) concept that similarity is greater than to predetermined threshold joins in initialized set, again expands to new term set; Described threshold value is the value calculated in repeatedly user's input, Multiple through then out similarity, and threshold value is excessive, the concept that very difficult expansion makes new advances; Threshold value is too little, and it is too large that the concept that expansion is come in and user input the semantic gap of keyword; The threshold value of the embodiment of the present invention gets 0.7.

Images