EP2689351A1

EP2689351A1 - Method for refining the results of a search within a database

Info

Publication number: EP2689351A1
Application number: EP12714812.0A
Authority: EP
Inventors: Eric Mathieu; Cyril MARCH
Original assignee: Xilopix SAS
Current assignee: Qwant SAS
Priority date: 2011-03-23
Filing date: 2012-03-19
Publication date: 2014-01-29
Also published as: US20140032544A1; RU2613039C2; EP3627353A1; CN103518198B; FR2973134A1; FR2973134B1; WO2012127168A1; RU2013146871A; KR20140027173A; CN103518198A; KR101980219B1

Abstract

The present invention relates to a method for refining the results of a search for objects in a database in which the objects are each related to descriptors, comprising the steps aiming to: - present a user with a set of objects associated with a means for the user to signal its relevance and/or its non-relevance with respect to his search - as a function of the signalling, assign at least one weight to the descriptors of an object presented considered by the user to be relevant and/or non-relevant, - calculate a resultant of the weight associated with each descriptor of the set of objects, - initialize a relevance index for each object, - compare each object, and for each descriptor of the object compared, modify the relevance index as a function of the weight of this descriptor in the resultant, - present the result objects in the order of their relevance index.

Description

Method for refining the results of a search in a database

The present invention relates to a method for refining the results of a search in a database containing a set of objects each linked to at least one descriptor.

The development of digital technologies in recent years, along with the development of networks and the Internet has led to an extremely large increase in the amount of digital content available.

A particularly significant example is the development of digital photography, particularly because of the development of websites for uploading and sharing photos. Thus, one of the leaders of this type of site has since September 2010 exceeded the five billion photos posted online and continues to put online several thousand per day.

These digital objects are generally listed in databases in association with keywords and / or other technical descriptors (size, resolution, etc.). These keywords and descriptors make it possible to search the database and return the objects whose keywords match the search criteria entered by a user in a search field.

However, currently, most search engines are primarily designed to search for text within Internet pages or files, and in particular in associated description texts.

In the case where the stored objects are not of a textual nature, such as photos, the associated keywords and descriptors are of considerable importance to enable efficient searching and return a relevant result.

Many search engines exist to carry out such research, and many algorithms have been developed to optimize the relevance of the results of this research.

Despite sophisticated algorithms, a keyword search has intrinsic limits, for example because of the existence in human language of synonyms, homonyms, hierarchy in terms, degree of precision. Within these limits, the precise search intent of the user beyond the original meaning of the keywords used remains unknown to the search engine.

To overcome these limitations, most search engines allow users to perform an advanced search, including using multiple keywords combined with each other by logical operators.

Such a search mode is however not particularly easy for the user and can, on some engines, go as far as requiring quasi-programming skills to write a request, while not knowing if this request can be correctly interpreted. by the motor and will lead to the desired result.

Thus, there is a need justifying the development of a method for optimizing the searches of objects contained in a database and in particular to overcome certain ambiguities or inaccuracies to better meet the request of the user.

To do this, the present invention relates to a method for refining the results of an object search in at least one database containing at least one set of objects each associated with at least one descriptor, said method comprising the steps to:

presenting to a user all or part of a set of objects in the database, at least some of the objects presented being each associated with at least one means for a user to indicate the relevance and / or less a means for the user to signal the irrelevance of said object in relation to his search,

according to the reports of the user, assign at least one po ids to all or part of the descriptors of an object of all the objects presented considered by the user to be relevant and / or irrelevant to his search ,

calculate a resultant of the weights associated with each descriptor of the set of result objects,

initialize an index of relevance for each result object, compare each result object to the resultant, and for each descriptor of the result object confronted, increase or decrease the relevance index of the object according to the weight of this descriptor in the resultant,

- present to the user all or part of the results objects in the order of their calculated relevance index.

Thus, by allowing the user to directly report the results of an initial search that he finds relevant or irrelevant, it is possible to better take into account the real meaning of his research and to provide a more satisfactory result. Moreover, thanks to such a method, it is easy for the user to carry out a complex search adding or removing descriptors and words and this, intu itive and transparent.

By object is meant any digital object that can be stored in a database. As stated above, it may include photos, but also audio files, video, documents, etc.

It should be noted that, in accordance with the operating principle of a database, the referenced objects are not necessarily contained directly in a database record and can be referenced by the database. intermediary of their storage address or URL, for example, or any other indirect means.

It should also be noted that the term descriptor used is not limited. The term descriptor obviously includes key word descriptors, but it can also be technical descriptors referencing textures, materials, color profiles, definition, etc. They can also be semantic descriptors established from a thesaurus. The nature of the descriptors is generally not limited and they can be adapted according to the objects referenced in the database or databases, and searched for.

It should also be noted that different weightings can be assigned to different descriptors, in particular according to their origin, context, situation with respect to all the other descriptors. Thus, for example, descriptors from a thesaurus, and therefore presenting a standardized, uniform and structured character, may have more weight than key-word descriptors that would have been attributed by the users themselves of a website. Photo.

Several unexpected and surprising beneficial effects have been noted. In particular, it appeared that the process according to the invention allowed to the user to overcome to a certain extent the language of the textual descriptors used. Indeed, from an initial search in his language, the user using the method according to the invention to refine the search results can also give weight transparently to descriptors and keywords in the language. alien associated with the object. In the end, research can be refined on the basis of foreign-language keywords, or at least by taking them into account, a foreign language that the user does not necessarily understand and therefore would not have entered directly into the foreign language. a textual search engine.

According to a preferred embodiment, the set of objects initially presented to the user corresponds to all or part of the objects resulting from a first search, in particular by keyword, in the database or databases. Of course, all initial search modes for generating a first set of objects are possible. In addition to a conventional search using a text field and a word entry by the user, one can imagine a selection of objects directly from geographic coordinates on a map, or a first photo that would, for example, be analyzed for extract search parameters, etc.

Depending on the number of objects returned by this initial search, we can choose to present to the user only part of the results, for example the first ten thousand photos of a search by keywords in a database photos.

It should also be noted that the search in the database or databases can be performed in an internal database, but also on external databases hosted on specialized remote sites for example.

One could also choose not to carry out a first search by keywords and to present to the user a set of standard objects representing major categories of the database, for example. The user is then free to browse the database by successively refining his selections using the method object of the present invention.

Preferably, the objects of the set of objects initially presented to the user are presented in a defined order when obtaining said set of objects, in particular an order of relevance by report to the initial search, this relevance being able to be defined by a search algorithm. In fact, traditional search engines often associate a relevance index with the results of their research.

Alternatively or in a complementary way, the order of relevance and initial presentation can be defined in an ad hoc way in order, for example, to maximize the number of different objects presented initially in order to leave as much choice as possible the user for his first refining, and possibly for the following ones.

According to a preferred embodiment, the weights assigned to the descriptors of the objects considered irrelevant is of opposite sign to the weights assigned to the descriptors of the objects considered to be relevant, and more particularly to signs respectively negative and positive.

Of course, it is simply a scale of notation given by way of example, the point of reference is not necessarily zero, other reference points can be chosen without difficulty and simply constituting a shift of time. 'ladder. In this case, the terms "opposite sign", "positive" and "negative" shall be considered to be understood in relation to this reference point.

According to a first variant, the absolute values of the weights assigned to the descriptors of the objects considered as relevant and / or irrelevant are equal.

According to a second variant, the weight assigned to the descriptors of the objects considered to be relevant has a different absolute value, and notably higher, than the weight assigned to the descriptors of the objects considered irrelevant.

Advantageously, the values of the weights assigned to the descriptors of the objects considered as relevant and / or irrelevant may be different for each object reported.

Advantageously, the value given to the descriptors of the objects considered relevant and / or irrelevant depends on their initial order of priority. In particular, a coefficient can be applied to a standard weight value. For example, an object considered 90% relevant by the search engine that performed the initial search may be assigned 90% of the reference weight value if that object is considered relevant by the user. In However, if the user considers it irrelevant, unlike the search engine, we can choose to allocate only 10% of the reference value of the irrelevant weight.

According to an advantageous embodiment, the means for signaling the relevance and / or the irrelevance of an object presented comprise means capable of indicating different degrees of relevance and / or irrelevance making it possible, in particular, to assign a different weight to depending on the degree of relevance and / or irrelevance reported. Thus, it may include a web page including buttons to indicate that an object is, for example, "very relevant" (first degree), "relevant" (second degree), "irrelevant" (third degree), "Irrelevant" (fourth degree) and "off topic" (fifth degree).

Advantageously, the result objects are presented in the form of previews, miniatures and / or extracts.

According to a particular embodiment, the objects contained in the database include photos and / or videos and / or audio objects. It may also be other types of documents, text files, etc.

According to a first variant embodiment, the relevance index is initialized to an identical value for each result object, in particular zero.

According to a second variant embodiment, the relevance index is initialized to different values for all or part of the result objects, in particular according to the initial order of presentation and, if appropriate, a relevance value returned by the search. initial.

According to a more advanced embodiment, all or some of the descriptors of the most relevant objects returned feed a new search in the database.

The present invention will be better understood in the light of the following detailed description with reference to the appended drawing in which:

FIG. 1 is a screenshot of an Internet site implementing the method according to the invention at the level of the first step presenting to a user the results of an initial search by keyword,

FIG. 2 is a screenshot of the site of FIG. 1 on which a user has reported a photo that he considers relevant to his search, FIG. 3 is a screenshot of the site of FIG. 1 on which a user has reported a photo that he considers irrelevant to his search,

FIG. 4 is a screen shot after initiation of the search refinement step by the user,

FIG. 5 is a screenshot of the site of FIG. 1 showing the result of the refining step carried out as a function of the relevance and irrelevance reports by the user,

FIG. 6 is a logic diagram showing schematically the flow of the process illustrated in FIGS. 1 to 5.

Referring also to Figure 6, Figures 1 to 5 show screenshots of an Internet site implementing the method according to the invention on a search for car headlight photos.

FIG. 1 shows a first step 110 in which a set of photo thumbnails P1 to P14 is presented to the user.

This set of photos P1 to P14 was obtained through an initial keyword search in one or more photo databases.

In this case, the key word "lighthouse" was used by the user to define his search and typed in a search field R of the page.

The search field R serves as an interface with the user and feeds a search engine, internal or external to the site, into photo databases. Such databases list a large number of photos and associate descriptors for subsequent searches. These descriptors include keyword lists, but can also be specific to the photo (camera used, technical data, color profile, etc.).

It is easy to understand that the use of a single key word "lighthouse" is naturally a source of ambiguity and carries different meanings that the search engine can not solve.

The search engine thus returns the results of its search algorithm and presents them to the user in the form of the fourteen thumbnails of photos P1 to P14.

It should be noted that the fourteen photos presented to the user do not necessarily correspond to the totality of the initial search results and it is quite possible to choose not to present to the user only part of the results, for example the first thousand photos returned.

As can be seen in FIG. 1, the photos P1, P2, P4, P5, P1, P8, P9, P1 1, P12 relate to photos of coastal lights for navigation.

Photos P3, P6, P1 0, P1 3, P14, meanwhile, refer to pictures of car headlights.

Each photo is associated, in its database or in another database, with one or more descriptors.

For the purposes of the example, we will consider that the photos P1, P2, P4, P5, P7, P8, P9, P1 1, P12 are associated with a "flagship" keyword descriptor, and that the photos P3, P6 , P10, P13, P14 are each associated with two "flagship" and "car" descriptors.

In accordance with the method according to the invention, the photos P1 to P14 are each presented to the user in association with a clickable image 11 representing a validation check mark and a clickable image 12 representing a prohibition cross.

These clickable images are associated with computer functions recording the user's choices and constitute means for said user to indicate the relevance (tick) and / or non-relevance (cross) of each photo in relation to his actual search.

It is obvious that the images representing a checkmark and a cross are given as an example and that any equivalent representation is possible, including clickable text informing the user of the choice he has.

The user then proceeds during a step 102 to report the photos that he considers relevant and / or irrelevant.

Figure 2 is a screenshot showing that the user has reported that picture P14 was relevant to his actual search. An M1 message informs him of the proper consideration of his report by the site or software.

Figure 3 is a screenshot showing that the user has reported that the P4 photo was not relevant to his actual search since it represents a coastal lighthouse. An M2 message informs him of the proper consideration of his report by the site or software.

In the present example, the messages M1 and M2 are displayed in the form of "pop-up" (display of a window in superposition). It is it is obvious that these messages can be signaled to the user in other forms, notably by grouping the selected images, displaying in a sidebar, setting up virtual baskets of the images selected as relevant and irrelevant, etc.

When the user has finished selecting the photos that he considers to be relevant and / or irrelevant to his search, he activates the process of refining the search by clicking, for example, on a button B. An example of treatment screen is shown in FIG. 4.

Of course, the refining process can also take place in real time depending on the interactions of the user, but this will require more processing and load resources of a remote server in particular.

The processing steps are transparent to the user.

During a step 103, a weight P is associated with each descriptor associated with each image signaled by the user. The weight P is negatively affected if the image has been reported as irrelevant and a positive sign if the image has been reported as relevant.

In the proposed example, photo P4, which has an associated "flagship" descriptor, was flagged irrelevant and photo P14, which has two associated "flagship" and "car" descriptors, was reported to be relevant.

Thus, the "flagship" descriptor is assigned a weight -P because of the irrelevance reported for photo P4 and is weighted + P because of the relevance reported for photo P14. Similarly, the "car" descriptor is weighted + P due to the relevance reported for photo P14.

A resultant of the weights assigned to each descriptor of the set of images P1 to P14 is computed during a step 104.

In this case, the "flagship" descriptor thus obtains a global weight of zero, while the "car" descriptor obtains an overall weight equal to + P.

The resultant is the set of descriptors of photos P1 to P4 assigned their respective weights previously calculated.

Before proceeding with the refining and sorting of the objects presented, a relevance index is associated with each photo P1 to P14 and initialized to zero during a step 105.

Each photo P1 to P14 therefore has the same priority and relevance. A step 1 06 makes it possible to compare each photo P1 to P14 with the resultant of the weights of the descriptors.

To do this, each descriptor of the picture P1 to P14 is compared to the resultant, and the priority index is increased or decreased by the weight of this descriptor in said resultant.

Thus, the photo P1, showing a coastal lighthouse, and possessing only the "flagship" descriptor, sees its priority index increased by the weight of the "flagship" descriptor in the resultant, namely zero. Its priority index remains at zero.

The same goes for photo P2.

The photo P3 on the other hand shows car headlights. As mentioned before, it is associated with two "flagship" and "car" descriptors. For the "flagship" descriptor, its index does not change, since the weight of this descriptor is zero. On the other hand, for its descriptor "car", its priority index is increased by the weight of the descriptor "car" in the resultant, namely + P. Its priority index becomes + P.

The same is true for photos P4 to P14.

It is therefore sufficient to rearrange the photos P1 to P14 according to their newly calculated priority index and to redisplay them in the order of their index of decreasing relevance during a step 1 07 so that the photos of car headlights are displayed first and then coastal lighthouses.

Figure 5 shows a screenshot showing the final rearrangement where only pictures of car headlights are well presented.

It will be seen that FIG. 5 also shows photos that were not present on the initial presentation screen. Indeed, it is quite possible to select a larger batch of photos than the fourteen pictures presented, some of which were taken by the user. However, they are present in the initial selection and taken into account for the implementation of the process. They therefore also receive a relevance index that changes their order in the selection. In the end, they can be found among the first fourteen photos, and thus be presented to the user.

The initial photos of lighthouses are, for their part, relegated beyond the fourteenth and therefore no longer appear. Of course, the user can then further refine his search, especially if new photos are presented to him (step 108) or stop his search (109).

Although the invention has been described with a particular embodiment, it is obvious that it is in no way limited and that it includes all the technical equivalents of the means described and their combinations if they enter into the scope of the invention.

In particular, it will be possible to provide additional signaling means, for example a "neutral" button in addition to means for signaling relevance and / or irrelevance.

It is also possible to provide a means of reinitializing the weights and relevance index in case of error by the user or if he wishes to begin refining according to other criteria.

Furthermore, although the present invention has been described with respect to photos, it is obviously not limited thereto and any other type of digital file to which descriptors may be associated may be used for its implementation. It will be possible to implement the process in the same way with audio files, especially associated with descriptors as to their musical style, their nature of sound, their instruments, etc., but also with video files, animated images , documents, text files, including scanned old books, etc.

Claims

1. A method for refining the results of an object search in at least one database containing at least one set of objects each associated with at least one descriptor, said method comprising the steps of:

presenting to a user all or part of a set of objects of the database, at least a portion of the objects presented being each associated with at least one means for a user to indicate the relevance and / or at least one means for the user to signal the irrelevance of said object in relation to his search,

- according to the reports of the user, assign at least a weight to all or part of the descriptors of an object of all the objects presented considered by the user to be relevant and / or irrelevant to his research ,

- initialize a relevance index for each result object,

- compare each result object to the resultant, and for each descriptor of the object confronted result, increase or decrease the index of relevance of the object according to the weight of this descriptor in the resultant,

2. Method according to claim 1, characterized in that the set of objects initially presented to the user corresponds to all or part of the objects resulting from a first search, in particular by keyword, in the database or databases .

3. Method according to any one of claims 1 or 2, characterized in that the objects of the set of objects initially presented to the user are presented in a defined order when obtaining said set of objects , in particular an order of relevance in relation to the initial search, this relevance being able to be defined in particular by a search algorithm.

4. Method according to any one of claims 1 to 3, characterized in that the weights assigned to the descriptors objects considered irrelevant are opposite to the weight assigned to the descriptors objects considered relevant, and more particularly signs respectively negative and positive.

5. Method according to any one of claims 1 to 4, characterized in that the absolute values of the weights assigned to the descriptors of the objects considered relevant and / or irrelevant are equal.

6. Method according to any one of claims 1 to 4, characterized in that the weight assigned to the descriptors of the objects considered relevant has a different absolute value, and especially higher, than the weight assigned to the descriptors of the objects considered as irrelevant.

7. Method according to any one of claims 1 to 6, characterized in that the values of the weights assigned to the descriptors of the objects considered relevant and / or irrelevant may be different for each object reported.

8. Method according to claim 7, characterized in that the weight value assigned to the descriptors of the objects considered relevant and / or irrelevant is a function of their initial order of priority.

9. Method according to any one of claims 1 to 8, characterized in that the means for signaling the relevance and / or irrelevance of an object presented comprise means capable of indicating different degrees of relevance and / or of not relevance allowing in particular to assign a different weight according to the degree of relevance and / or irrelevance reported.

10. Method according to any one of claims 1 to 9, characterized in that the result objects are presented in the form of previews, thumbnails and / or extracts.

1 1. Method according to one of Claims 1 to 10, characterized in that the objects contained in the database comprise photos and / or videos and / or audio objects.

12. Method according to any one of claims 1 to 10, characterized in that the relevance index is initialized to an identical value for each result object, including zero.

13. Method according to any one of claims 1 to 1 0, characterized in that the relevance index is initialized to different values for all or part of the result objects, in particular according to the initial order of presentation and if applicable the relevance value returned by the initial search.

14. Method according to any one of claims 1 to 13, characterized in that all or part of the descriptors of the most relevant objects returned feed a new search in the database.