WO2018042204A1

WO2018042204A1 - Method of displaying search results

Info

Publication number: WO2018042204A1
Application number: PCT/GB2017/052572
Authority: WO
Inventors: Adam EASTON
Original assignee: SimCentric Limited
Priority date: 2016-09-05
Filing date: 2017-09-05
Publication date: 2018-03-08
Also published as: GB201615038D0; GB2553360A

Abstract

A method of analysing search results using a computing device having at least a processor, a memory, a display device and an input device, comprises : a) enabling a user to enter into said memory at least two search terms using said input device; b) using said at least two search terms and said processor to perform a search of a data set and to provide one or more search results, being the results of said search; c) displaying on said display device: i) a search term particle corresponding to each search term; and ii) a search result particle corresponding to each search result; d) positioning each search term particle at a different position on said display device; e) using said processor to calculate for each search result particle: i) a force of attraction between the search result particle and at least one of the search term particles; and ii) a force of repulsion between the search result particle and at least some other search result particles; and f) displaying said search result particles on said display device in positions which correspond with the balance of said forces of attraction and repulsion.

Description

METHOD OF DISPLAYING SEARCH RESULTS

FIELD OF THE INVENTION This invention relates to searching technology. BACKGROUND OF THE INVENTION

Existing search technologies have a number of limitations. Advances in search technologies over the last two decades have focussed primarily on algorithms relating to indexing the data and searching through that index. Over this period of significant advances in underlying search algorithms, several elements have remained largely unchanged. The first of these limitations, is that nearly all search algorithms currently return a static list of search results. Search algorithms all currently return an ordered list of results, with some algorithms providing an additional search score related to each result which dictates how closely that search result matches the search term. The second limitation of existing search algorithms is that they are not interactive. Once a results list is retrieved and displayed, the user must enter a new search to retrieve a new list. This means that once a search result is revealed to a user, there is no easy way to interact with those search results other than viewing them. The third limitation of existing search algorithms is the way that they deal with multiple search terms. Generally, search algorithms will combine multiple search terms/phrases so that the presence of any of the terms will increase the search score of the relevant search score. This in effect provides an aggregate score based on all the search terms entered within a single query. What this approach does not provide is an understanding of how much each search term within the query contributed to the overall search score, or whether two search results which may be returned with similar scores have relevance to similar search terms within the query.

The consequence of these limitations in the current approach to search is that a user's ability to use search algorithms to explore relationships within a data set is limited. SUMMARY OF THE INVENTION

The invention provides a method, apparatus, programmed computer and computer- readable medium, as set out in the accompanying claims.

BRIEF DESCRIPTION OF THE FIGURES

Figure 1 shows three search term particles and a number of search result particles displayed on a display device, together with a search interface for entering search terms;

Figure 2 shows the search term particles and search result particles of Figure 1 in greater detail;

Figure 3 is a flow diagram for the process of adding a search term;

Figure 4 is a flow diagram for a simulation loop which generates forces on the particles, and updates the states of the particles;

Figure 5 is an example of results on a display device for three search terms A, B and C;

Figure 6 is an example of results on a display device in which a number of search result particles are divided into groups along a one-dimensional axis representing a variable;

Figure 7 shows the results of Figure 6 where search term particles representing the search terms A, B and C are arranged in a line perpendicular to the one-dimensional axis, and the search term particles are divided according to the search terms A, B and C; and

Figure 8 shows a computing device suitable for carrying out methods described herein.

DESCRIPTION OF PREFERRED EMBODIMENTS Preferred embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings.

We describe an apparatus, method and algorithm for searching data which allows a human user to more quickly identify individual pieces of data or key clusters of data or similar information, as well as identifying how information relates to different search terms.

The embodiments described below may be performed using dedicated hardware, or may be performed using software on a computing device, or using any combination of hardware and software. It will therefore be appreciated that the components shown in the accompanying drawings may represent hardware modules in the case of a dedicated hardware implementation, or may represent steps in a method, which may for example be performed in software.

Figure 8 shows a computing device 60, which may for example be a personal computer (PC), on which methods described herein can be carried out. The computing device 60 comprises a display 62 for displaying information, a processor 64, a memory 68 and an input device 70 for allowing information to be input to the computing device. The input device 70 may for example include a connection to other computers or to computer readable media, and may also include a mouse or keyboard for allowing a user to enter information. These elements are connected by a bus 72 via which information is exchanged between the components. A preferred method has: a. The potential to be interactive in nature;

b. The ability to display information in a multi-dimensional space rather than as a simple ordered list;

c. The ability to deal with large numbers of search terms simultaneously; and d. The capability of working with small data sets right up to very large data sets.

The method relies on the concept of treating each search term and each search result as a particle within a multi-dimensional space. Each such particle may be displayed on the display 62 of the computing device 60 shown in Figure 8. Figure 1 shows an example in which three Search Term Particles 2 are displayed, together with a number of Search Result Particles 4, each representing a result of the search.

These particles 2, 4 are then moved on the display 62 by a combination of forces which are derived from: a. Relationships between search terms and search results;

b. The current physical state (location) of the particles in the world;

c. User interaction; and/or

d. Search world boundaries.

In a preferred method, every particle 2, 4 has a force exerted on it from one or more force generators, which provide the forces described above. These forces are then summed within each of a series of time intervals, referred to as time steps, and each particle 2, 4 is accelerated based on the total force acting on the particle 2, 4 within each time step.

The result is a spatial representation of the search results and search terms. An example in two-dimensional space is shown in Figure 1. A search box 6 is provided on display 62 in which three search terms 8 (Truck, Large and Car) have been entered by a user. Each search result is represented by a Search Result Particle 4, which may be arranged in circles or ovals 10 around each Search Term Particle 2, as shown in Figure 1 . As shown in Figure 1 , a number of Search Result Particles 4 may cluster around each Search Term Particle. Separate clusters 1 1 of Search Result Particles 4 may form in separate regions between each pair of Search Term Particles 2. This allows the user to effectively filter the search results according to which search terms are most relevant to each result. For example, the cluster 1 1 between the Search Term Particles 2 for "Large" and "Truck" are search results which are relevant to, or contain, "Large" and "Truck", but which are not relevant to, or do not contain, "Car". Thus the results are not simply presented to the user as a list, but are arranged in a way which allows filtering and analysis of the data. Particle Forces

As noted above, there are two different types of particles that are created in the method, these are Search Term Particles 2 and Search Result Particles 4. Each type has a different set of forces applied to them. In Figure 2, the three search terms (Truck, Car, and Large) can be seen as distinct Search Term Particles 2 whereas all of the smaller Search Result Particles 4 move and cluster around them. Search Term Particles

These particles 2 are affected by forces arising from user interactions (such as dragging or moving the particles 2 manually) or forces attracting the Search Term Particles 2 to pre-calculated positions within the multi-dimensional space. The Search Term Particle state, that is the position of the Search Term Particles 2, is not affected by the Search Result Particles 4.

Search Result Particles

Search Result Particles 4 move much more freely than the Search Term Particles 2. They are affected by Search Term Particles 2 as well as by other close-by Search Result Particles 4. They may also be affected by any boundaries of the multidimensional space. Generally, when a Search Term Particle 2 is moved (by user interaction), this will cause all Search Result Particles 4 to move accordingly.

Algorithm Flow

There are two execution paths as part of the algorithm. The first is executed each time a new search term is entered. The second is executed from within a simulation loop. Adding a Search Term

The algorithm flow diagram for the process of adding a Search Term is shown in Figure 3. The major components of this process are as follows: a. Search Term Entry

Referring to Figure 3, in the Search Term Entry module or step 12, search terms 14 are entered into the system, for example using the input device 70 of Figure 8. These can be any type of data accepted by the Search Algorithm 16, however, in the example of Figure 2, the search terms are text. In the case of a text Search Term, a single Search Term could contain multiple words. The Search Term Entry could be generated by user input from a text field or automatically from another program or process. b. Search Algorithm

The Search Algorithm 16 represents a component akin to current commonly used searching tools. The input of the module is a search term 14, and the output is an ordered list of search results 18. Depending on the search algorithm used, these search results may have a search score associated with each result. This search score would indicate how closely the search result matched the search term. Other search algorithms may not produce a search score, but rely on the order of the outputted list to define relative matching to the search term. c. Search Score Normalisation

The search normalisation module or step 20 takes the search results from the search algorithm 16 and adjusts the search scores so that they fall within a reasonable range. A suitable algorithm for the search score normalisation 20 is:

1 . Input normalisation parameters;

2. Input ordered list of search results 18;

a. If search results do not include a search score generated by the search algorithm 16, allocate these;

3. Loop through each search term;

a. Determine mean score for all search terms;

b. Determine maximum score of all search terms;

c. Determine minimum score of all search terms;

d. Determine standard deviation of all search term scores;

4. Determine desired normalisation mapping equation based on calculated score details and normalisation parameters;

5. Loop through each search term;

a. Adjust search term score based on mapping from actual score to desired score;

6. Output ordered list of search terms with these adjusted scores 22;

The normalisation parameters can include details like:

a. Desired mean

b. Desired maximum value

c. Desired minimum value

d. Desired standard deviation

The mapping equation in step 4 can be derived via a number of established methods from mapping one range of numbers to another range of numbers.

The normalisation process is important to stop certain search results from dominating other search results. Depending on the output of search results, it is possible that certain search terms will have scores orders of magnitude higher than others. When, in subsequent steps of the algorithm, these scores are converted to Force vectors, large differences in search scores can result in non-intuitive behaviours of the particles 2, 4. An additional advantage of the normalisation process is that it can be tweaked to affect the clustering of the search particles 2, 4. For example, in the case where a set of Search Results are directly relevant to two Search Terms that have been entered, normalising the search results to a range with a low standard deviation will result in a tight cluster of search results which relate to both search terms. As the standard deviation is increased the cluster is elongated along the line between the two search terms so that the search results appear in more of a continuum according to their relative relevance to each search term. d. Particle Converter

The Particle Converter module or step 24 is responsible for converting the Search Result data into particle data. If the same Search Result is returned by more than one Search Term (eg. the Search Result is related to both Search Terms), then only one Particle will be created. As such, each particle represents a discreet piece of information.

As particles are generated they are saved to the Particle Set Internal Storage 26. This ensures that a record of all existing particles are maintained so that for each Search Result processed, it can be checked whether to create a new particle or update an existing one.

Each particle 2, 4 may store any or all of the following internal data:

a. Particle state, which includes:

a. Position (N dimensions)

b. Velocity (N dimensions)

c. Acceleration (N dimensions)

d. Notional mass (N dimensions)

b. Search Result data

c. Search Output data, which is a set one or more results, where each result includes:

a. Link to the associated Search term b. Relevant (normalised) Search score

A suitable algorithm for the Particle Converter 24 (which may be run for each Search term 14 at the time it is added) is:

a. For each Search Result from this Search Term

i. Check Particle Set Internal Storage for existing Particle with the same Search Result

1 . If matching particle found

a. Update existing particle with new record showing the search score and related Search term for this term

i. Add additional search output data entry for particle

2. Else

a. Create new particle and populate it with this Search Term data.

Initialise state data

Set search result data

Add first search output data entry for particle

b. Add new particle to Particle Set Internal Storage

Within step a.i.2.a.i, the step of Initialising state data can be seen. There are several strategies that can be employed. These include (but are not limited to) initialising particle positions at the edge of the world boundaries, randomly initialising particle positions, or adding particles according some other placement algorithm. e. Filter

The Filter module or step 28 may remove particles from the set based on a series of defined Filter rules. This allows a user to quickly remove from the output of the algorithm any results that do not fit desired criteria. In this module filters can be applied based on the properties of the particle 'state' (such as particle position or particle velocity etc) as well as based on the properties of the corresponding Search Result data. Examples of Filters that can be applied are:

a. Only showing particles whose Search Results contain a particular word; b. Removing particles whose Search Results contain a date field later than a particular time; and/or

c. Removing a set of particles that have been selected manually by the user (for example through a lasso tool or region highlighting on the display 62).

A suitable algorithm for the Filter module 28 is:

a. Input is set of Filters and set of Particles;

b. Create empty output set of Particles;

c. For each Filter

a. For each Particle in input set

i. Evaluate Filter criteria against particle

1 . If Filter returns true

a. Add particle to output set

2. Else

a. Do nothing

d. Return Output set of particles

Simulation Loop

A suitable algorithm flow diagram for a simulation loop is shown in Figure 4. This process may be run as a continuous loop. This allows the algorithm to react to user induced changes in a responsive manner. f. Force Generator

Referring to Figure 4, a force generator module or step 30 is responsible for determining the individual force vectors which affect the particles 2, 4. There can be any number of forces acting on the particles. In this embodiment there is a force exerted on Search Result Particles 4 by Search Term Particles 2 if the search result of the Search Result Particle 4 corresponds with the search term of the Search Term Particle 2. In this embodiment there is also a repulsive force working on Search Result Particles 4 which increases the closer they get to any other particles 2, 4.

The force vectors acting on the particles are vectors of a dimension corresponding to the dimensionality of the particle representation.

Suitable equations to determine the force vectors as a function of distance include: A constant force over distance; or

A linearly changing force over distance.

Regardless of the equation chosen, the magnitude of the force for each particle can related to the normalised search score between the search term and search result.

The balance of forces is important for the emergent properties of the system. Tuning of the force parameters can occur during the implementation of this algorithm to ensure that particles are organised in a useful way.

A recommended set of Forces applying to each Search Result Particle 4 is described below:

a. Search Result Particle

a. Attraction to relevant Search Terms

i. F = s ^* c ^* w ^* u , where F is the force vector, s is the normalised search score, c is a constant scaling factor, w is an option weight specified by the user for each Search Term to make some search terms stronger relative to others, and ύ is the unit vector pointing from the Search Term to the Search Particle

b. Repulsion from nearby Particles

i. F = ( c ^* s ^* w/d^A2 + b) * u, where F is the force vector, s is the normalised search score, c is a constant scaling factor, w is an option weight specified by the user for each Search Term to make some search terms stronger relative to others, d is distance between the particles, b is a constant and u is the unit vector pointing from the Search Term to the Search Particles

For computational efficiency when finding nearby Particles, a quad tree, or similar spatial organisation algorithm can be deployed in order to manage complexity. This is particularly important for search problems where large numbers of forces are returned.

A suitable algorithm for the Force Generator 30 is:

1 . For each particle

a. If particle is a Search Result Particle 4

i. For each Search Term Particle 2 1 . Generate and store Force vector b. For each force

i. Sum vector = sum vector + this force vector c. Store sum vector g. Force Summation

The force summation module or step 32 sums each of the individual force vectors acting on each particle, in order to create a single resultant force acting on each particle.

A suitable algorithm is:

2. For each particle

a. Create sum vector (initialised to 0 vector)

b. For each force

i. Sum vector = sum vector + this force vector c. Store sum vector h. Particle State Updater

The particle state updater 34 loops through each particle and uses the calculated combined Force to determine an acceleration vector.

A suitable algorithm for this module is:

1 . For each particle

a. Calculate acceleration vector based on Force vector and mass (a = F/m) b. Calculate velocity vector based on update timestep (deltaT) and acceleration

c. Calculate position based on timestep, velocity and acceleration d. Update acceleration, velocity and position for particle

When updating the particles on the simulation step, only the Particle state is updated, all other Particle fields are unchanged.

Analysing Particles in separate Lower Dimensions The power of the particle representation of Search Results can be further exploited by using different attractive forces at lower dimensions than the overall dimensionality of the particle. For example, considering a 2D particle, forces can be represented as a one-dimensional (1 D) force along the x axis and a 1 D force along the y axis. Therefore, rather than calculating the Force vector as a unit vector between Search Term and Search Result, two axis aligned, orthogonal unit vectors can be used to allow further analysis of the results.

Figure 5 shows an example of how the results of a search (using 3 search terms A, B and C) may look using the standard algorithm described in the previous sections. However, the user may wish to look at how the results look when analysed with respect to a continuous variable within the source data. An example of such a continuous variable could include time/date, monetary amount, quantity, measurement etc. In this case, maintaining the particle representation and by using a different variation of the Force Generator module, a self-organising histogram organisation of the data can be achieved. To do this, a one-dimensional force, aligned with the x axis is calculated. This force pushes the particles towards their correct position along the continuous variable line.

Concurrently, another force is used to push the particles in a one-dimensional direction aligned to the Y axis. This force is pushing the particles down towards the 0 value of that axis. The particles then separate based on the mutual repulsion of the close- together particles, forcing late arriving particles further away from the X axis.

Figure 6 shows a particle distribution achievable with this approach.

As shown in Figure 7, this approach can be further enhanced by adding search terms to this representation. In this case, the Search Term Particles 2 can be represented as a one-dimensional stack 40. Wherever they are placed in that stack, they will attract search terms in that 1 D space to the correct level. This then allows a dynamic representation showing the relationship between the continuous variable (for example time along the x-axis 42) and the discreet search terms A, B and C, represented by the Search Term Particles 2 in the stack 40. Figure 7 shows an example of this using a timeline as the continuous variable. The approach described above scales to higher dimensions as well, so that for three- dimensional (3D) particles, i.e. particles represented in a representation of three- dimensional space (which may be displayed on a 2D display), a continuous variable could be represented along a one-dimensional (1 D) axis, with the particles free to move based on search terms in the remaining two dimensions.

Each feature disclosed or illustrated in the present specification may be incorporated in the invention, whether alone or in any appropriate combination with any other feature disclosed or illustrated herein.

The embodiments described thus provide a method for exploring data by representing generated search results as force-directed particles. This allows representing one or more ordered lists of search results as a set of force-directed particles where the force between those particles is derived from their order in the list. This also allows representing one or more ordered lists of search results as a set of force-directed particles where the force between those particles is derived from their respective search scores, and representing search terms as force directed pins which will then exert forces on the search result particles. The previous three representations may be added to a continuously updating loop in order to build an interactive and responsive virtual environment containing the search results. The user can thus fuse the outputs of multiple discreet search terms into a single representation where the interrelations between these discreet searches can be analysed. This allows the ability to filter the Search Result particles based on the 'state' of those search result particles, rather than the content of the Search Result themselves. Separate treatment of the dimensions within a higher dimensional representation is possible, allowing individual properties of the search results to exert independent 1 D forces on the particles. This allows a seamless transition between the different methods of organising data within this particle framework.

Claims

CLAIMS:

1. A method of analysing search results using a computing device having at least a processor, a memory, a display device and an input device, said method comprising: a) enabling a user to enter into said memory at least two search terms using said input device;

b) using said at least two search terms and said processor to perform a search of a data set and to provide one or more search results, being the results of said search;

c) displaying on said display device:

i) a search term particle corresponding to each search term; and ii) a search result particle corresponding to each search result;

d) positioning each search term particle at a different position on said display device;

e) using said processor to calculate for each search result particle:

i) a force of attraction between the search result particle and at least one of the search term particles; and

ii) a force of repulsion between the search result particle and at least some other search result particles; and

f) displaying said search result particles on said display device in positions which correspond with the balance of said forces of attraction and repulsion.

2. A method as claimed in claim 1 , which further comprises:

for each search result calculating a search result score which represents the degree to which the search result matches a corresponding one of said search terms; and

wherein said force of attraction between a search result particle and a search term particle is calculated to be greater where the search score result between the corresponding search result and search term is greater, and is calculated to be lower where the search score result between the corresponding search result and search term is lower.

3. A method as claimed in claim 2, which further comprises normalising said search result scores to ensure that the search result scores fall within a predetermined range.

4. A method as claimed in any preceding claim, wherein said force of repulsion between said search result particles increases as the distance between said search result particles decreases.

5. A method as claimed in any preceding claim, which further comprises allowing said user to move each search term particle on said display device.

6. A method as claimed in claim 5, wherein when a search term particle is moved by a user, at least some of said search result particles also move to new positions on said display device as a result of changes in said forces.

7. A method as claimed in any preceding claim, wherein said forces of attraction and repulsion cause a number of search result particles to cluster around each search term particle.

8. A method as claimed in claim 7, wherein said search result particles cluster around each search term particle in a generally circular or oval configuration.

9. A method as claimed in any preceding claim, wherein said forces of attraction and repulsion cause a number of search result particles to cluster in separate regions between each pair of search term particles.

10. A method as claimed in claim 9, wherein search result particles in one of said separate regions between a pair of search term particles correspond with search results which are relevant to, or contain, the two search terms corresponding with said pair of search term particles.

11 . A method as claimed in any preceding claim, wherein said forces of attraction and repulsion are summed within each of a series of time intervals, and said search result particles move on said display device so that each search result particle is accelerated based on the total force acting on the particle within each time step.

12. A method as claimed in any preceding claim, which further comprises:

after performing the method of any preceding claim, allowing said user to enter into said memory a further search term, and repeating the method of any preceding claim using said at least two search terms and said further search term.

13. A method as claimed in any preceding claim, which further comprises for each search result particle storing in said memory a particle state which includes any or all of the following:

i) the position of the particle;

ii) the velocity of the particle;

iii) the acceleration of the particle; and

iv) a notional mass of the particle.

14. A method as claimed in any preceding claim, which further comprises filtering said search result particles to remove search result particles which do not meet certain specified criteria.

15. A method as claimed in any preceding claim, which further comprises representing said search term particles and search result particles, and their position and movement, in three dimensions on said display device.

16. A method as claimed in any preceding claim, which further comprises providing on said display device a representation of a variable along an axis in one dimension, and positioning said search result particles along lines or regions corresponding with different values along said axis.

17. A method as claimed in claim 16, which further comprises arranging said search term particles in a line, or one-dimensional stack, adjacent said axis.

18 Search analysis apparatus comprising at least a processor, a memory, a display device and an input device, wherein said apparatus is arranged to perform a method as claimed in any preceding claim.

19. A programmed computer comprising at least a processor, a memory, a display device and an input device, wherein said computer is programmed to perform a method as claimed in any one of claims 1 to 17.

20. A computer-readable medium containing computer-readable instructions for performing a method as claimed in any one of claims 1 to 17.