Fig. 1. Past and current query similarities. Note: The actual query can be better processed by using knowledge that is encapsulated in ten available past queries. The shape and fill colour of the graphical representation of queries illustrates a semantic similarity between them, where similar queries have similar shapes and colours.

Fig. 2. Selection of different important features included in current query.

Fig. 3. Intelligent exploration using similar queries.

Fig. 4. Similar query neighbourhood and its unique representation as expected query Q^(e).

Fig. 5. Adaptation of actual query Q^(a) by moving it towards centroid Q^(e).

Fig. 6. Adaptation of Q^(a) depends on presence or absence of each attribute in Q^(a).

Fig. 7. Component structure of manager agent responsible for performing exploration activities.

Fig. 8. Combined distance functions with different d_E and d_J weights.

Fig. 9. Procedure used for evaluating different exploration strategies.

List of 10 past queries

Note: Attribute values and weights are marked as value/weight. Columns d_E ≡ d_E(Q^(a), Q⁽ⁱ⁾), d_J ≡ d_J(Q^(a), Q⁽ⁱ⁾), o_S ≡ o_S(Q^(a), Q⁽ⁱ⁾) and d_C ≡ d_C(Q^(a), Q⁽ⁱ⁾) correspond to the query Q^(a)(a₁(3, 0.5), a₇(4, 0.2), a₈(3, 0.4)).

Values of the mean m and standard deviation σ for the attributes

Expected attribute value-importance pairs for neighbourhoods in Table 1 for (k = 3, 5, 10)

Note: The columns correspond to attributes from the actual query Q^(a)(a₁(3, 0.5), a₇(4,0.2), a₈(3, 0.4)).

Average scores for combined distance function and best manually-achievable score

Note: The results are for any query in Set S_A.

Distribution of user feedback values

Note: These results are based on nine different configurations of the proposed exploratory approach, as well as the collaborative filtering (CF), and the mutation and crossover (MC) approaches.

Actual queries adapted by adding five “Most Promising” words

Note: The results presented in the table were found for a query similarity neighbourhood of size 3. The words that appeared in both the neighbourhood, as well as the adapted actual query are shown in bold face.