Employee Searching Based On User Computer Science Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

The growth of the Web and the Internet leads to the development of an ever increasing number of interesting application classes. The most common method used now in companies is normal recruitment process. If a company wants an employee immediately, the only way for recruitment is advertising in any media. After receiving applications from the employees, they need to check the qualification, experience etc. It is a time required process.

This paper proposes a method for employee searching by using a user and query dependent ranking. In this paper we present a ranking model based on user inputs. This ranking model is acquired from several other ranking functions derived for various user-query pairs. This is based on the intuition that similar users display comparable ranking preferences over the result of similar queries. This paper gives an idea about how the ranking can be used.

Keywords- User Similarity, Query Similarity, Automatic Ranking, Workload, Relational Queries.


The success and growth of the Internet and Web leads to the development of a large number of Web databases for a variety of applications. Database systems support only a Boolean query model. If query is not selective then too many tuples may be in the answer. It is time consuming to select the most appropriate answer .Web databases simplify this task by sorting the query result. Currently this sorting is done on the values of a single attribute. The ordering based on multiple attribute values would be closer to the Web user's expectation.

We use the following two scenarios as our running examples.

Example-1: Two users - a software company executive (U1) and a nonsoftware company executive, for example a data entry company (U2), seek answers to the same query (Q1): "Working area= computer AND Location = Dallas, TX", for which more than 18,000 tuples are typically returned in response. Intuitively, U1would typically search for employees with Programming skills in particular language, and hence would prefer employees with "Condition =programmer AND language = Java" tobe ranked and displayed higher than the others. In contrast, U2 would most likely search for data entry operators with minimum speed in data entry; hence, for U2, employees with "Condition = Dataentry operator AND qualification=Plus Two" should be displayed before the rest.

Example-2: The same user (U2) moves to do some medical transcription work and asks a different query (say Q4): "Working area = Medical field AND Location = Mountain View". We can presume that he may want employees with slightly higher qualification for medical transcription, and hence would prefer employees with "Condition = Data Entry Operator AND Qualification=Degree" to be ranked higher than others.

Example-1 shows that towards the results of the same query, different Web Users may have contrasting ranking preferences [2]. Example-2 shows that the same user may display different ranking preferences for the results of different queries [2]. Thus in the case of Web databases, where a large set of queries is involved, the corresponding results should be ranked in a user-and query-dependent manner.

The current sorting mechanism used by Web databases is an automated ranking of database results. Automated ranking provide a single ranking order for a given query across all users because they do not differentiate between users. In contrast, techniques for building extensive user profiles [3] as well as requiring users to order data tuples [4], proposed for user-dependent ranking, do not distinguish between queries and provide a single ranking order for any query given by the same user.

In this paper, we propose an application of user- and query-dependent approach for ranking the results of Web databases queries. The key goal of an information retrieval system is to retrieve information which might be useful or relevant to the user. Employees are recruited into the company by normal methods such campus placements, advertising in any media etc. But it is a time required process. For filling a single vacancy the above method is not as efficient. For that purpose we propose a method for employee searching by using a user and query dependent ranking. The employer can search in the site and can select employees with required qualification.

For a query Qj given by a user Ui, a relevant ranking function is identified from a workload of ranking functions, to rank Qj's results. Query similarity indicates that for the results of a given query, similar users display comparable ranking preferences. And the user similarity means a user displays analogous ranking preferences over results of similar queries. The ranking function we used is a function of attribute weights and value weights. The former denoting the significance of individual attributes and the latter representing the importance of attribute values. A minimal workload is important to make our approach practically useful. By adapting relevant feedback mechanism, we can acquire such a workload.

Related work

There was no concept of ranking in traditional databases. Currently ranking has become everywhere at once and is used in document retrieval systems, traditional data bases, Web searching/browsing as well.

Ranking done in database

This context proposes address the problem of query dependent ranking. But, for a given query, this technique provides the same ordering of tuples across all users. By considering the profiles of users for user-dependent ranking in databases has been proposed in. A drawback in all these works is that they do not consider that the same user may have varied ranking preferences for different queries. The closest form of query- and user-dependent ranking in relational databases has been proposed in. This technique is also unsuitable for Web users who are not proficient with query languages and ranking functions. In contrast, our framework provides an automated query- as well as user-dependent ranking solution without requiring users to possess knowledge about query languages, data models and ranking mechanisms.

Relevance Feedback

Inferring a ranking function by analyzing the user's interaction with the query results originates from the concepts of relevance feedback [7] [8] [9] in the domain of document and image retrieval systems. The direct application of either explicit or implicit feedback mechanisms for inferring database ranking functions has several challenges.

problem definition and architecture

The ranking problem can be stated as: "For the query Qjgiven by the user Ui, determine a ranking function FUiQjfrom W".The ranking problem can be split into:

1. Identifying a ranking function using the similarity model: Given W, determine a user Uxsimilar to Uiand a query Qysimilar to Qjsuch that the function FUxQyexists in W.

2. Generating a workload of ranking functions: Given a user Uxasking query Qy, based on Ux's preferences towards Qy's results, determine, explicitly or implicitly, a ranking function FUxQy.W is then established as a collection of such ranking functions learnt over different user-query pairs.

A. Ranking Architecture

The core component of ranking framework is the similarity model(Figure 1). The set of users ({Ui, U1, U2, ...Ur}) most similar to Ui, determined by the user similarity model

C:\Users\vishu\Desktop\similarity.jpgFig. 1. Similarity Ranking Model [2]

The query similarity model determines the set of queries ({Qj,Q1,Q2, ...,Qp}) most similar to Qj. Using these similar queries and users, it searches the workload to identify the functionFUxQy. The ranking functions for several user-query pairs are formed from the workload used in our framework.

Our ranking function is of the linear weighted-sum type. The mechanism used for deriving this function captures the: i) significance associated by the user to each attribute i.e., an attribute-weight and ii) user's emphasis on individual values of an attribute i.e., a value-weight.


Where wi represents the attribute-weight of Ai and vi representsthe value-weight for Ai's value in tuple t.

similarity model for ranking

When ranking functions are known for a small set of user-query pairs, then the concept of similarity-based ranking is aimed. At the time of answering a query asked by a user, if no ranking function is available for this user-query pair, the proposed query and user-similarity models can effectively identify a suitable function to rank the corresponding results.

Query Similarity

For the user U1 from Example-1, a ranking function does not exist for ranking Q1's results(N1). However, from Example-2, we know that a user is likely to have displayed different ranking preferences for different query results. Consequently, a randomly selected function from U1's workload is not likely to give a desirable ranking order over N1. On the other hand, the ranking functions are likely to be comparable for queries similar to each other [2].

We advance the hypothesis that if Q1 is most similar to query Qy (in U1's workload), U1 would display similar ranking preferences over the results of both queries; thus, the ranking function (F1y) derived for Qy can be used to rank N1. Similar to recommendation systems, our framework can utilize the aggregate function, composed from the functions corresponding to the top-k most similar queries to Q1, to rank N1 [2]. We translate this proposal of query similarity into two alternative models: i) query condition similarity, and ii) query-result similarity.

A.1. Query-Condition Similarity

By comparing the attribute values in the query conditions, the similarity between two queries can be determined.

Given two queries Q and Q', each with theconjunctive selection conditions, respectively of the form"WHERE A1=a1 AND · · · AND Am=am" and "WHEREA1=a1' AND · · · AND Am=am' " , the query-condition similarity betweenQ and Q'is given as the conjunctive similarities between thevalues ai and ai' for every attribute Ai (Equation 1).

Similarity(Q,Q') =sim(Q[Ai = ai],Q'[Ai = ai'])(2)

A.2. Query-Result Similarity

If two queries are similar, the results are likely to greater similarity. Similaritybetween a pair of queries is valuated as the similarity between the tuples in the respective query results. Given two queries Q and Q', let N and N'betheir query results. The query-result similarity between Q and

Q'is then computed as the similarity between the result setsN and N', given by Equation 2.

similarity(Q,Q') =sim(N,N') (3)


Fig 2. Query similarity model summarized view

The above figure shows the computation of similarity for the two models.

User Similarity

We know from Example-1 that different users may display different ranking preferences towards the same query. We put forward the hypothesis that if U1is similar toan existing user Ux, then, for the results of a given query (say Q1), both users will show similar ranking preferences; therefore, Ux's ranking function (Fx1) can be used to rank Q1's results for U1 as well. Given two users Ui and Uj with the set of common queries - {Q1, Q2, ..., Qr}, for which ranking functions ({Fi1, Fi2, ..., Fir} and {Fj1, Fj2, ..., Fjr}) exist in W, the user similarity between Ui and Uj is expressed as the average similarity between their individual ranking functions for each query Qp (shown in Equation 3):

Similarity(Ui, Uj) (4)

c. The Composite Similarity Model

The goal of this composite model is to determine a ranking function (Fxy). Finding such an appropriate ranking function is given by the Algorithm.

INPUT:Ui, Qj , Workload W (M queries, N users)

OUTPUT: Ranking Function Fxy to be used for Ui, Qj


for p = 1 to M do

%% Using Equation 2 %%

Calculate Query Condition Similarity (Qj ,Qp)

end for

%% Based on descending order of similarity with Qj %%

Sort(Q1, Q2, .... QM)

Select QKset i.e., top-K queries from the above sorted set


for r = 1 to N do

%% Using Equation 4 %%

Calculate User Similarity (Ui, Ur) over QKset

end for

%% Based on descending order of similarity with Ui %%

Sort(U1, U2, .... UN) to yield Uset


for Each Qs ∈QKset do

for Each Ut∈Uset do

Rank(Ut,Qs) =Rank(Ut∈Uset) + Rank(Qs ∈QKset)

end for

end for

Fxy = Get-RankingFunction()

The input to the algorithm is a user (Ui) and a query

(Qj) along with the workload matrix (W) containing ranking

functions. The algorithm begins by determining the querycondition similarity (STEP ONE) between Qj and every query in the workload. It then sorts all these queries (in descending order) based on their similarity with Qj and selects the set (QKset) of the top-K most similar queries to Qj that satisfy the conditions for the top-K user similarity model. Based on these selected queries, the algorithm determines the usersimilarity(STEP TWO) between Ui and every user in theworkload. All the users are then sorted (again, in descendingorder) based on their similarity to Ui. We then generate a listof all the user-query pairs (by combining the elements fromthe two sorted sets), and linearise these pairs by assigning arank (which is the sum of query and use similarity ranks) toeach pair (STEP THREE). For instance, if Ux and Qy occuras the xth and yth elements in the respective ordering with theinput pair, the pair (Ux, Qy) are assigned an aggregate rank.In this case, a rank of "x + y" will be assigned. The "Get-RankingFunction" method then selects the pair (Ux, Qy) thathas the lowest combined rank and contains a ranking function

(Fxy) in the workload. Then, in order to rank the results (Nj ), the corresponding attribute weights and value weights btained

forFxy will be individually applied to each tuple in Nj.

Workload of ranking functions

In this paper, the proposed model uses a workload of ranking functions. Obtaining such a ranking function is not a trivial task in the context of Web databases.Since obtainingranking functions from users on the Web is difficult determining the exact set of ranking functions to be derived for establishing the workload is important.


In this paper, we propose a model for employee searching by using a user- and query-dependent ranking method.By using this method, it solves the Many-Answers Problem which leverages data and workload statistics and correlations.The design and maintenance of an appropriate workload that satisfies properties of similarity-based ranking is very challenging.


We thank the anonymous referees for their extremely useful comments on an earlier draft of this article.